Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/page_alloc.c | |
3 | * | |
4 | * Manages the free list, the system allocates free pages here. | |
5 | * Note that kmalloc() lives in slab.c | |
6 | * | |
7 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
8 | * Swap reorganised 29.12.95, Stephen Tweedie | |
9 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 | |
10 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 | |
11 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 | |
12 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 | |
13 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 | |
14 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) | |
15 | */ | |
16 | ||
17 | #include <linux/config.h> | |
18 | #include <linux/stddef.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/swap.h> | |
21 | #include <linux/interrupt.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/bootmem.h> | |
24 | #include <linux/compiler.h> | |
9f158333 | 25 | #include <linux/kernel.h> |
1da177e4 LT |
26 | #include <linux/module.h> |
27 | #include <linux/suspend.h> | |
28 | #include <linux/pagevec.h> | |
29 | #include <linux/blkdev.h> | |
30 | #include <linux/slab.h> | |
31 | #include <linux/notifier.h> | |
32 | #include <linux/topology.h> | |
33 | #include <linux/sysctl.h> | |
34 | #include <linux/cpu.h> | |
35 | #include <linux/cpuset.h> | |
bdc8cb98 | 36 | #include <linux/memory_hotplug.h> |
1da177e4 LT |
37 | #include <linux/nodemask.h> |
38 | #include <linux/vmalloc.h> | |
4be38e35 | 39 | #include <linux/mempolicy.h> |
1da177e4 LT |
40 | |
41 | #include <asm/tlbflush.h> | |
42 | #include "internal.h" | |
43 | ||
44 | /* | |
45 | * MCD - HACK: Find somewhere to initialize this EARLY, or make this | |
46 | * initializer cleaner | |
47 | */ | |
c3d8c141 | 48 | nodemask_t node_online_map __read_mostly = { { [0] = 1UL } }; |
7223a93a | 49 | EXPORT_SYMBOL(node_online_map); |
c3d8c141 | 50 | nodemask_t node_possible_map __read_mostly = NODE_MASK_ALL; |
7223a93a | 51 | EXPORT_SYMBOL(node_possible_map); |
6c231b7b RT |
52 | unsigned long totalram_pages __read_mostly; |
53 | unsigned long totalhigh_pages __read_mostly; | |
1da177e4 | 54 | long nr_swap_pages; |
8ad4b1fb | 55 | int percpu_pagelist_fraction; |
1da177e4 | 56 | |
d98c7a09 | 57 | static void __free_pages_ok(struct page *page, unsigned int order); |
a226f6c8 | 58 | |
1da177e4 LT |
59 | /* |
60 | * results with 256, 32 in the lowmem_reserve sysctl: | |
61 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) | |
62 | * 1G machine -> (16M dma, 784M normal, 224M high) | |
63 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA | |
64 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL | |
65 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA | |
a2f1b424 AK |
66 | * |
67 | * TBD: should special case ZONE_DMA32 machines here - in those we normally | |
68 | * don't need any ZONE_NORMAL reservation | |
1da177e4 | 69 | */ |
a2f1b424 | 70 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { 256, 256, 32 }; |
1da177e4 LT |
71 | |
72 | EXPORT_SYMBOL(totalram_pages); | |
1da177e4 LT |
73 | |
74 | /* | |
75 | * Used by page_zone() to look up the address of the struct zone whose | |
76 | * id is encoded in the upper bits of page->flags | |
77 | */ | |
c3d8c141 | 78 | struct zone *zone_table[1 << ZONETABLE_SHIFT] __read_mostly; |
1da177e4 LT |
79 | EXPORT_SYMBOL(zone_table); |
80 | ||
a2f1b424 | 81 | static char *zone_names[MAX_NR_ZONES] = { "DMA", "DMA32", "Normal", "HighMem" }; |
1da177e4 LT |
82 | int min_free_kbytes = 1024; |
83 | ||
84 | unsigned long __initdata nr_kernel_pages; | |
85 | unsigned long __initdata nr_all_pages; | |
86 | ||
13e7444b | 87 | #ifdef CONFIG_DEBUG_VM |
c6a57e19 | 88 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
1da177e4 | 89 | { |
bdc8cb98 DH |
90 | int ret = 0; |
91 | unsigned seq; | |
92 | unsigned long pfn = page_to_pfn(page); | |
c6a57e19 | 93 | |
bdc8cb98 DH |
94 | do { |
95 | seq = zone_span_seqbegin(zone); | |
96 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) | |
97 | ret = 1; | |
98 | else if (pfn < zone->zone_start_pfn) | |
99 | ret = 1; | |
100 | } while (zone_span_seqretry(zone, seq)); | |
101 | ||
102 | return ret; | |
c6a57e19 DH |
103 | } |
104 | ||
105 | static int page_is_consistent(struct zone *zone, struct page *page) | |
106 | { | |
1da177e4 LT |
107 | #ifdef CONFIG_HOLES_IN_ZONE |
108 | if (!pfn_valid(page_to_pfn(page))) | |
c6a57e19 | 109 | return 0; |
1da177e4 LT |
110 | #endif |
111 | if (zone != page_zone(page)) | |
c6a57e19 DH |
112 | return 0; |
113 | ||
114 | return 1; | |
115 | } | |
116 | /* | |
117 | * Temporary debugging check for pages not lying within a given zone. | |
118 | */ | |
119 | static int bad_range(struct zone *zone, struct page *page) | |
120 | { | |
121 | if (page_outside_zone_boundaries(zone, page)) | |
1da177e4 | 122 | return 1; |
c6a57e19 DH |
123 | if (!page_is_consistent(zone, page)) |
124 | return 1; | |
125 | ||
1da177e4 LT |
126 | return 0; |
127 | } | |
128 | ||
13e7444b NP |
129 | #else |
130 | static inline int bad_range(struct zone *zone, struct page *page) | |
131 | { | |
132 | return 0; | |
133 | } | |
134 | #endif | |
135 | ||
224abf92 | 136 | static void bad_page(struct page *page) |
1da177e4 | 137 | { |
224abf92 | 138 | printk(KERN_EMERG "Bad page state in process '%s'\n" |
7365f3d1 HD |
139 | KERN_EMERG "page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n" |
140 | KERN_EMERG "Trying to fix it up, but a reboot is needed\n" | |
141 | KERN_EMERG "Backtrace:\n", | |
224abf92 NP |
142 | current->comm, page, (int)(2*sizeof(unsigned long)), |
143 | (unsigned long)page->flags, page->mapping, | |
144 | page_mapcount(page), page_count(page)); | |
1da177e4 | 145 | dump_stack(); |
334795ec HD |
146 | page->flags &= ~(1 << PG_lru | |
147 | 1 << PG_private | | |
1da177e4 | 148 | 1 << PG_locked | |
1da177e4 LT |
149 | 1 << PG_active | |
150 | 1 << PG_dirty | | |
334795ec HD |
151 | 1 << PG_reclaim | |
152 | 1 << PG_slab | | |
1da177e4 | 153 | 1 << PG_swapcache | |
689bcebf | 154 | 1 << PG_writeback ); |
1da177e4 LT |
155 | set_page_count(page, 0); |
156 | reset_page_mapcount(page); | |
157 | page->mapping = NULL; | |
9f158333 | 158 | add_taint(TAINT_BAD_PAGE); |
1da177e4 LT |
159 | } |
160 | ||
1da177e4 LT |
161 | /* |
162 | * Higher-order pages are called "compound pages". They are structured thusly: | |
163 | * | |
164 | * The first PAGE_SIZE page is called the "head page". | |
165 | * | |
166 | * The remaining PAGE_SIZE pages are called "tail pages". | |
167 | * | |
168 | * All pages have PG_compound set. All pages have their ->private pointing at | |
169 | * the head page (even the head page has this). | |
170 | * | |
41d78ba5 HD |
171 | * The first tail page's ->lru.next holds the address of the compound page's |
172 | * put_page() function. Its ->lru.prev holds the order of allocation. | |
173 | * This usage means that zero-order pages may not be compound. | |
1da177e4 | 174 | */ |
d98c7a09 HD |
175 | |
176 | static void free_compound_page(struct page *page) | |
177 | { | |
178 | __free_pages_ok(page, (unsigned long)page[1].lru.prev); | |
179 | } | |
180 | ||
1da177e4 LT |
181 | static void prep_compound_page(struct page *page, unsigned long order) |
182 | { | |
183 | int i; | |
184 | int nr_pages = 1 << order; | |
185 | ||
d98c7a09 | 186 | page[1].lru.next = (void *)free_compound_page; /* set dtor */ |
41d78ba5 | 187 | page[1].lru.prev = (void *)order; |
1da177e4 LT |
188 | for (i = 0; i < nr_pages; i++) { |
189 | struct page *p = page + i; | |
190 | ||
5e9dace8 | 191 | __SetPageCompound(p); |
4c21e2f2 | 192 | set_page_private(p, (unsigned long)page); |
1da177e4 LT |
193 | } |
194 | } | |
195 | ||
196 | static void destroy_compound_page(struct page *page, unsigned long order) | |
197 | { | |
198 | int i; | |
199 | int nr_pages = 1 << order; | |
200 | ||
41d78ba5 | 201 | if (unlikely((unsigned long)page[1].lru.prev != order)) |
224abf92 | 202 | bad_page(page); |
1da177e4 LT |
203 | |
204 | for (i = 0; i < nr_pages; i++) { | |
205 | struct page *p = page + i; | |
206 | ||
224abf92 NP |
207 | if (unlikely(!PageCompound(p) | |
208 | (page_private(p) != (unsigned long)page))) | |
209 | bad_page(page); | |
5e9dace8 | 210 | __ClearPageCompound(p); |
1da177e4 LT |
211 | } |
212 | } | |
1da177e4 | 213 | |
17cf4406 NP |
214 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
215 | { | |
216 | int i; | |
217 | ||
218 | BUG_ON((gfp_flags & (__GFP_WAIT | __GFP_HIGHMEM)) == __GFP_HIGHMEM); | |
6626c5d5 AM |
219 | /* |
220 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO | |
221 | * and __GFP_HIGHMEM from hard or soft interrupt context. | |
222 | */ | |
223 | BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); | |
17cf4406 NP |
224 | for (i = 0; i < (1 << order); i++) |
225 | clear_highpage(page + i); | |
226 | } | |
227 | ||
1da177e4 LT |
228 | /* |
229 | * function for dealing with page's order in buddy system. | |
230 | * zone->lock is already acquired when we use these. | |
231 | * So, we don't need atomic page->flags operations here. | |
232 | */ | |
233 | static inline unsigned long page_order(struct page *page) { | |
4c21e2f2 | 234 | return page_private(page); |
1da177e4 LT |
235 | } |
236 | ||
237 | static inline void set_page_order(struct page *page, int order) { | |
4c21e2f2 | 238 | set_page_private(page, order); |
1da177e4 LT |
239 | __SetPagePrivate(page); |
240 | } | |
241 | ||
242 | static inline void rmv_page_order(struct page *page) | |
243 | { | |
244 | __ClearPagePrivate(page); | |
4c21e2f2 | 245 | set_page_private(page, 0); |
1da177e4 LT |
246 | } |
247 | ||
248 | /* | |
249 | * Locate the struct page for both the matching buddy in our | |
250 | * pair (buddy1) and the combined O(n+1) page they form (page). | |
251 | * | |
252 | * 1) Any buddy B1 will have an order O twin B2 which satisfies | |
253 | * the following equation: | |
254 | * B2 = B1 ^ (1 << O) | |
255 | * For example, if the starting buddy (buddy2) is #8 its order | |
256 | * 1 buddy is #10: | |
257 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 | |
258 | * | |
259 | * 2) Any buddy B will have an order O+1 parent P which | |
260 | * satisfies the following equation: | |
261 | * P = B & ~(1 << O) | |
262 | * | |
263 | * Assumption: *_mem_map is contigious at least up to MAX_ORDER | |
264 | */ | |
265 | static inline struct page * | |
266 | __page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order) | |
267 | { | |
268 | unsigned long buddy_idx = page_idx ^ (1 << order); | |
269 | ||
270 | return page + (buddy_idx - page_idx); | |
271 | } | |
272 | ||
273 | static inline unsigned long | |
274 | __find_combined_index(unsigned long page_idx, unsigned int order) | |
275 | { | |
276 | return (page_idx & ~(1 << order)); | |
277 | } | |
278 | ||
279 | /* | |
280 | * This function checks whether a page is free && is the buddy | |
281 | * we can do coalesce a page and its buddy if | |
13e7444b NP |
282 | * (a) the buddy is not in a hole && |
283 | * (b) the buddy is free && | |
284 | * (c) the buddy is on the buddy system && | |
285 | * (d) a page and its buddy have the same order. | |
4c21e2f2 | 286 | * for recording page's order, we use page_private(page) and PG_private. |
1da177e4 LT |
287 | * |
288 | */ | |
289 | static inline int page_is_buddy(struct page *page, int order) | |
290 | { | |
13e7444b NP |
291 | #ifdef CONFIG_HOLES_IN_ZONE |
292 | if (!pfn_valid(page_to_pfn(page))) | |
293 | return 0; | |
294 | #endif | |
295 | ||
1da177e4 LT |
296 | if (PagePrivate(page) && |
297 | (page_order(page) == order) && | |
1da177e4 LT |
298 | page_count(page) == 0) |
299 | return 1; | |
300 | return 0; | |
301 | } | |
302 | ||
303 | /* | |
304 | * Freeing function for a buddy system allocator. | |
305 | * | |
306 | * The concept of a buddy system is to maintain direct-mapped table | |
307 | * (containing bit values) for memory blocks of various "orders". | |
308 | * The bottom level table contains the map for the smallest allocatable | |
309 | * units of memory (here, pages), and each level above it describes | |
310 | * pairs of units from the levels below, hence, "buddies". | |
311 | * At a high level, all that happens here is marking the table entry | |
312 | * at the bottom level available, and propagating the changes upward | |
313 | * as necessary, plus some accounting needed to play nicely with other | |
314 | * parts of the VM system. | |
315 | * At each level, we keep a list of pages, which are heads of continuous | |
316 | * free pages of length of (1 << order) and marked with PG_Private.Page's | |
4c21e2f2 | 317 | * order is recorded in page_private(page) field. |
1da177e4 LT |
318 | * So when we are allocating or freeing one, we can derive the state of the |
319 | * other. That is, if we allocate a small block, and both were | |
320 | * free, the remainder of the region must be split into blocks. | |
321 | * If a block is freed, and its buddy is also free, then this | |
322 | * triggers coalescing into a block of larger size. | |
323 | * | |
324 | * -- wli | |
325 | */ | |
326 | ||
48db57f8 | 327 | static inline void __free_one_page(struct page *page, |
1da177e4 LT |
328 | struct zone *zone, unsigned int order) |
329 | { | |
330 | unsigned long page_idx; | |
331 | int order_size = 1 << order; | |
332 | ||
224abf92 | 333 | if (unlikely(PageCompound(page))) |
1da177e4 LT |
334 | destroy_compound_page(page, order); |
335 | ||
336 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); | |
337 | ||
338 | BUG_ON(page_idx & (order_size - 1)); | |
339 | BUG_ON(bad_range(zone, page)); | |
340 | ||
341 | zone->free_pages += order_size; | |
342 | while (order < MAX_ORDER-1) { | |
343 | unsigned long combined_idx; | |
344 | struct free_area *area; | |
345 | struct page *buddy; | |
346 | ||
1da177e4 | 347 | buddy = __page_find_buddy(page, page_idx, order); |
1da177e4 LT |
348 | if (!page_is_buddy(buddy, order)) |
349 | break; /* Move the buddy up one level. */ | |
13e7444b | 350 | |
1da177e4 LT |
351 | list_del(&buddy->lru); |
352 | area = zone->free_area + order; | |
353 | area->nr_free--; | |
354 | rmv_page_order(buddy); | |
13e7444b | 355 | combined_idx = __find_combined_index(page_idx, order); |
1da177e4 LT |
356 | page = page + (combined_idx - page_idx); |
357 | page_idx = combined_idx; | |
358 | order++; | |
359 | } | |
360 | set_page_order(page, order); | |
361 | list_add(&page->lru, &zone->free_area[order].free_list); | |
362 | zone->free_area[order].nr_free++; | |
363 | } | |
364 | ||
224abf92 | 365 | static inline int free_pages_check(struct page *page) |
1da177e4 | 366 | { |
92be2e33 NP |
367 | if (unlikely(page_mapcount(page) | |
368 | (page->mapping != NULL) | | |
369 | (page_count(page) != 0) | | |
1da177e4 LT |
370 | (page->flags & ( |
371 | 1 << PG_lru | | |
372 | 1 << PG_private | | |
373 | 1 << PG_locked | | |
374 | 1 << PG_active | | |
375 | 1 << PG_reclaim | | |
376 | 1 << PG_slab | | |
377 | 1 << PG_swapcache | | |
b5810039 | 378 | 1 << PG_writeback | |
92be2e33 | 379 | 1 << PG_reserved )))) |
224abf92 | 380 | bad_page(page); |
1da177e4 | 381 | if (PageDirty(page)) |
242e5468 | 382 | __ClearPageDirty(page); |
689bcebf HD |
383 | /* |
384 | * For now, we report if PG_reserved was found set, but do not | |
385 | * clear it, and do not free the page. But we shall soon need | |
386 | * to do more, for when the ZERO_PAGE count wraps negative. | |
387 | */ | |
388 | return PageReserved(page); | |
1da177e4 LT |
389 | } |
390 | ||
391 | /* | |
392 | * Frees a list of pages. | |
393 | * Assumes all pages on list are in same zone, and of same order. | |
207f36ee | 394 | * count is the number of pages to free. |
1da177e4 LT |
395 | * |
396 | * If the zone was previously in an "all pages pinned" state then look to | |
397 | * see if this freeing clears that state. | |
398 | * | |
399 | * And clear the zone's pages_scanned counter, to hold off the "all pages are | |
400 | * pinned" detection logic. | |
401 | */ | |
48db57f8 NP |
402 | static void free_pages_bulk(struct zone *zone, int count, |
403 | struct list_head *list, int order) | |
1da177e4 | 404 | { |
c54ad30c | 405 | spin_lock(&zone->lock); |
1da177e4 LT |
406 | zone->all_unreclaimable = 0; |
407 | zone->pages_scanned = 0; | |
48db57f8 NP |
408 | while (count--) { |
409 | struct page *page; | |
410 | ||
411 | BUG_ON(list_empty(list)); | |
1da177e4 | 412 | page = list_entry(list->prev, struct page, lru); |
48db57f8 | 413 | /* have to delete it as __free_one_page list manipulates */ |
1da177e4 | 414 | list_del(&page->lru); |
48db57f8 | 415 | __free_one_page(page, zone, order); |
1da177e4 | 416 | } |
c54ad30c | 417 | spin_unlock(&zone->lock); |
1da177e4 LT |
418 | } |
419 | ||
48db57f8 | 420 | static void free_one_page(struct zone *zone, struct page *page, int order) |
1da177e4 LT |
421 | { |
422 | LIST_HEAD(list); | |
48db57f8 NP |
423 | list_add(&page->lru, &list); |
424 | free_pages_bulk(zone, 1, &list, order); | |
425 | } | |
426 | ||
427 | static void __free_pages_ok(struct page *page, unsigned int order) | |
428 | { | |
429 | unsigned long flags; | |
1da177e4 | 430 | int i; |
689bcebf | 431 | int reserved = 0; |
1da177e4 LT |
432 | |
433 | arch_free_page(page, order); | |
de5097c2 IM |
434 | if (!PageHighMem(page)) |
435 | mutex_debug_check_no_locks_freed(page_address(page), | |
a4fc7ab1 | 436 | PAGE_SIZE<<order); |
1da177e4 | 437 | |
1da177e4 | 438 | for (i = 0 ; i < (1 << order) ; ++i) |
224abf92 | 439 | reserved += free_pages_check(page + i); |
689bcebf HD |
440 | if (reserved) |
441 | return; | |
442 | ||
48db57f8 | 443 | kernel_map_pages(page, 1 << order, 0); |
c54ad30c | 444 | local_irq_save(flags); |
a74609fa | 445 | __mod_page_state(pgfree, 1 << order); |
48db57f8 | 446 | free_one_page(page_zone(page), page, order); |
c54ad30c | 447 | local_irq_restore(flags); |
1da177e4 LT |
448 | } |
449 | ||
a226f6c8 DH |
450 | /* |
451 | * permit the bootmem allocator to evade page validation on high-order frees | |
452 | */ | |
453 | void fastcall __init __free_pages_bootmem(struct page *page, unsigned int order) | |
454 | { | |
455 | if (order == 0) { | |
456 | __ClearPageReserved(page); | |
457 | set_page_count(page, 0); | |
7835e98b | 458 | set_page_refcounted(page); |
545b1ea9 | 459 | __free_page(page); |
a226f6c8 | 460 | } else { |
a226f6c8 DH |
461 | int loop; |
462 | ||
545b1ea9 | 463 | prefetchw(page); |
a226f6c8 DH |
464 | for (loop = 0; loop < BITS_PER_LONG; loop++) { |
465 | struct page *p = &page[loop]; | |
466 | ||
545b1ea9 NP |
467 | if (loop + 1 < BITS_PER_LONG) |
468 | prefetchw(p + 1); | |
a226f6c8 DH |
469 | __ClearPageReserved(p); |
470 | set_page_count(p, 0); | |
471 | } | |
472 | ||
7835e98b | 473 | set_page_refcounted(page); |
545b1ea9 | 474 | __free_pages(page, order); |
a226f6c8 DH |
475 | } |
476 | } | |
477 | ||
1da177e4 LT |
478 | |
479 | /* | |
480 | * The order of subdivision here is critical for the IO subsystem. | |
481 | * Please do not alter this order without good reasons and regression | |
482 | * testing. Specifically, as large blocks of memory are subdivided, | |
483 | * the order in which smaller blocks are delivered depends on the order | |
484 | * they're subdivided in this function. This is the primary factor | |
485 | * influencing the order in which pages are delivered to the IO | |
486 | * subsystem according to empirical testing, and this is also justified | |
487 | * by considering the behavior of a buddy system containing a single | |
488 | * large block of memory acted on by a series of small allocations. | |
489 | * This behavior is a critical factor in sglist merging's success. | |
490 | * | |
491 | * -- wli | |
492 | */ | |
085cc7d5 | 493 | static inline void expand(struct zone *zone, struct page *page, |
1da177e4 LT |
494 | int low, int high, struct free_area *area) |
495 | { | |
496 | unsigned long size = 1 << high; | |
497 | ||
498 | while (high > low) { | |
499 | area--; | |
500 | high--; | |
501 | size >>= 1; | |
502 | BUG_ON(bad_range(zone, &page[size])); | |
503 | list_add(&page[size].lru, &area->free_list); | |
504 | area->nr_free++; | |
505 | set_page_order(&page[size], high); | |
506 | } | |
1da177e4 LT |
507 | } |
508 | ||
1da177e4 LT |
509 | /* |
510 | * This page is about to be returned from the page allocator | |
511 | */ | |
17cf4406 | 512 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
1da177e4 | 513 | { |
92be2e33 NP |
514 | if (unlikely(page_mapcount(page) | |
515 | (page->mapping != NULL) | | |
516 | (page_count(page) != 0) | | |
334795ec HD |
517 | (page->flags & ( |
518 | 1 << PG_lru | | |
1da177e4 LT |
519 | 1 << PG_private | |
520 | 1 << PG_locked | | |
1da177e4 LT |
521 | 1 << PG_active | |
522 | 1 << PG_dirty | | |
523 | 1 << PG_reclaim | | |
334795ec | 524 | 1 << PG_slab | |
1da177e4 | 525 | 1 << PG_swapcache | |
b5810039 | 526 | 1 << PG_writeback | |
92be2e33 | 527 | 1 << PG_reserved )))) |
224abf92 | 528 | bad_page(page); |
1da177e4 | 529 | |
689bcebf HD |
530 | /* |
531 | * For now, we report if PG_reserved was found set, but do not | |
532 | * clear it, and do not allocate the page: as a safety net. | |
533 | */ | |
534 | if (PageReserved(page)) | |
535 | return 1; | |
536 | ||
1da177e4 LT |
537 | page->flags &= ~(1 << PG_uptodate | 1 << PG_error | |
538 | 1 << PG_referenced | 1 << PG_arch_1 | | |
539 | 1 << PG_checked | 1 << PG_mappedtodisk); | |
4c21e2f2 | 540 | set_page_private(page, 0); |
7835e98b | 541 | set_page_refcounted(page); |
1da177e4 | 542 | kernel_map_pages(page, 1 << order, 1); |
17cf4406 NP |
543 | |
544 | if (gfp_flags & __GFP_ZERO) | |
545 | prep_zero_page(page, order, gfp_flags); | |
546 | ||
547 | if (order && (gfp_flags & __GFP_COMP)) | |
548 | prep_compound_page(page, order); | |
549 | ||
689bcebf | 550 | return 0; |
1da177e4 LT |
551 | } |
552 | ||
553 | /* | |
554 | * Do the hard work of removing an element from the buddy allocator. | |
555 | * Call me with the zone->lock already held. | |
556 | */ | |
557 | static struct page *__rmqueue(struct zone *zone, unsigned int order) | |
558 | { | |
559 | struct free_area * area; | |
560 | unsigned int current_order; | |
561 | struct page *page; | |
562 | ||
563 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { | |
564 | area = zone->free_area + current_order; | |
565 | if (list_empty(&area->free_list)) | |
566 | continue; | |
567 | ||
568 | page = list_entry(area->free_list.next, struct page, lru); | |
569 | list_del(&page->lru); | |
570 | rmv_page_order(page); | |
571 | area->nr_free--; | |
572 | zone->free_pages -= 1UL << order; | |
085cc7d5 NP |
573 | expand(zone, page, order, current_order, area); |
574 | return page; | |
1da177e4 LT |
575 | } |
576 | ||
577 | return NULL; | |
578 | } | |
579 | ||
580 | /* | |
581 | * Obtain a specified number of elements from the buddy allocator, all under | |
582 | * a single hold of the lock, for efficiency. Add them to the supplied list. | |
583 | * Returns the number of new pages which were placed at *list. | |
584 | */ | |
585 | static int rmqueue_bulk(struct zone *zone, unsigned int order, | |
586 | unsigned long count, struct list_head *list) | |
587 | { | |
1da177e4 | 588 | int i; |
1da177e4 | 589 | |
c54ad30c | 590 | spin_lock(&zone->lock); |
1da177e4 | 591 | for (i = 0; i < count; ++i) { |
085cc7d5 NP |
592 | struct page *page = __rmqueue(zone, order); |
593 | if (unlikely(page == NULL)) | |
1da177e4 | 594 | break; |
1da177e4 LT |
595 | list_add_tail(&page->lru, list); |
596 | } | |
c54ad30c | 597 | spin_unlock(&zone->lock); |
085cc7d5 | 598 | return i; |
1da177e4 LT |
599 | } |
600 | ||
4ae7c039 | 601 | #ifdef CONFIG_NUMA |
8fce4d8e CL |
602 | /* |
603 | * Called from the slab reaper to drain pagesets on a particular node that | |
604 | * belong to the currently executing processor. | |
879336c3 CL |
605 | * Note that this function must be called with the thread pinned to |
606 | * a single processor. | |
8fce4d8e CL |
607 | */ |
608 | void drain_node_pages(int nodeid) | |
4ae7c039 | 609 | { |
8fce4d8e | 610 | int i, z; |
4ae7c039 CL |
611 | unsigned long flags; |
612 | ||
8fce4d8e CL |
613 | for (z = 0; z < MAX_NR_ZONES; z++) { |
614 | struct zone *zone = NODE_DATA(nodeid)->node_zones + z; | |
4ae7c039 CL |
615 | struct per_cpu_pageset *pset; |
616 | ||
23316bc8 | 617 | pset = zone_pcp(zone, smp_processor_id()); |
4ae7c039 CL |
618 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
619 | struct per_cpu_pages *pcp; | |
620 | ||
621 | pcp = &pset->pcp[i]; | |
879336c3 CL |
622 | if (pcp->count) { |
623 | local_irq_save(flags); | |
624 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); | |
625 | pcp->count = 0; | |
626 | local_irq_restore(flags); | |
627 | } | |
4ae7c039 CL |
628 | } |
629 | } | |
4ae7c039 CL |
630 | } |
631 | #endif | |
632 | ||
1da177e4 LT |
633 | #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU) |
634 | static void __drain_pages(unsigned int cpu) | |
635 | { | |
c54ad30c | 636 | unsigned long flags; |
1da177e4 LT |
637 | struct zone *zone; |
638 | int i; | |
639 | ||
640 | for_each_zone(zone) { | |
641 | struct per_cpu_pageset *pset; | |
642 | ||
e7c8d5c9 | 643 | pset = zone_pcp(zone, cpu); |
1da177e4 LT |
644 | for (i = 0; i < ARRAY_SIZE(pset->pcp); i++) { |
645 | struct per_cpu_pages *pcp; | |
646 | ||
647 | pcp = &pset->pcp[i]; | |
c54ad30c | 648 | local_irq_save(flags); |
48db57f8 NP |
649 | free_pages_bulk(zone, pcp->count, &pcp->list, 0); |
650 | pcp->count = 0; | |
c54ad30c | 651 | local_irq_restore(flags); |
1da177e4 LT |
652 | } |
653 | } | |
654 | } | |
655 | #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */ | |
656 | ||
657 | #ifdef CONFIG_PM | |
658 | ||
659 | void mark_free_pages(struct zone *zone) | |
660 | { | |
661 | unsigned long zone_pfn, flags; | |
662 | int order; | |
663 | struct list_head *curr; | |
664 | ||
665 | if (!zone->spanned_pages) | |
666 | return; | |
667 | ||
668 | spin_lock_irqsave(&zone->lock, flags); | |
669 | for (zone_pfn = 0; zone_pfn < zone->spanned_pages; ++zone_pfn) | |
670 | ClearPageNosaveFree(pfn_to_page(zone_pfn + zone->zone_start_pfn)); | |
671 | ||
672 | for (order = MAX_ORDER - 1; order >= 0; --order) | |
673 | list_for_each(curr, &zone->free_area[order].free_list) { | |
674 | unsigned long start_pfn, i; | |
675 | ||
676 | start_pfn = page_to_pfn(list_entry(curr, struct page, lru)); | |
677 | ||
678 | for (i=0; i < (1<<order); i++) | |
679 | SetPageNosaveFree(pfn_to_page(start_pfn+i)); | |
680 | } | |
681 | spin_unlock_irqrestore(&zone->lock, flags); | |
682 | } | |
683 | ||
684 | /* | |
685 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. | |
686 | */ | |
687 | void drain_local_pages(void) | |
688 | { | |
689 | unsigned long flags; | |
690 | ||
691 | local_irq_save(flags); | |
692 | __drain_pages(smp_processor_id()); | |
693 | local_irq_restore(flags); | |
694 | } | |
695 | #endif /* CONFIG_PM */ | |
696 | ||
a74609fa | 697 | static void zone_statistics(struct zonelist *zonelist, struct zone *z, int cpu) |
1da177e4 LT |
698 | { |
699 | #ifdef CONFIG_NUMA | |
1da177e4 LT |
700 | pg_data_t *pg = z->zone_pgdat; |
701 | pg_data_t *orig = zonelist->zones[0]->zone_pgdat; | |
702 | struct per_cpu_pageset *p; | |
703 | ||
a74609fa | 704 | p = zone_pcp(z, cpu); |
1da177e4 | 705 | if (pg == orig) { |
e7c8d5c9 | 706 | p->numa_hit++; |
1da177e4 LT |
707 | } else { |
708 | p->numa_miss++; | |
e7c8d5c9 | 709 | zone_pcp(zonelist->zones[0], cpu)->numa_foreign++; |
1da177e4 LT |
710 | } |
711 | if (pg == NODE_DATA(numa_node_id())) | |
712 | p->local_node++; | |
713 | else | |
714 | p->other_node++; | |
1da177e4 LT |
715 | #endif |
716 | } | |
717 | ||
718 | /* | |
719 | * Free a 0-order page | |
720 | */ | |
1da177e4 LT |
721 | static void fastcall free_hot_cold_page(struct page *page, int cold) |
722 | { | |
723 | struct zone *zone = page_zone(page); | |
724 | struct per_cpu_pages *pcp; | |
725 | unsigned long flags; | |
726 | ||
727 | arch_free_page(page, 0); | |
728 | ||
1da177e4 LT |
729 | if (PageAnon(page)) |
730 | page->mapping = NULL; | |
224abf92 | 731 | if (free_pages_check(page)) |
689bcebf HD |
732 | return; |
733 | ||
689bcebf HD |
734 | kernel_map_pages(page, 1, 0); |
735 | ||
e7c8d5c9 | 736 | pcp = &zone_pcp(zone, get_cpu())->pcp[cold]; |
1da177e4 | 737 | local_irq_save(flags); |
a74609fa | 738 | __inc_page_state(pgfree); |
1da177e4 LT |
739 | list_add(&page->lru, &pcp->list); |
740 | pcp->count++; | |
48db57f8 NP |
741 | if (pcp->count >= pcp->high) { |
742 | free_pages_bulk(zone, pcp->batch, &pcp->list, 0); | |
743 | pcp->count -= pcp->batch; | |
744 | } | |
1da177e4 LT |
745 | local_irq_restore(flags); |
746 | put_cpu(); | |
747 | } | |
748 | ||
749 | void fastcall free_hot_page(struct page *page) | |
750 | { | |
751 | free_hot_cold_page(page, 0); | |
752 | } | |
753 | ||
754 | void fastcall free_cold_page(struct page *page) | |
755 | { | |
756 | free_hot_cold_page(page, 1); | |
757 | } | |
758 | ||
8dfcc9ba NP |
759 | /* |
760 | * split_page takes a non-compound higher-order page, and splits it into | |
761 | * n (1<<order) sub-pages: page[0..n] | |
762 | * Each sub-page must be freed individually. | |
763 | * | |
764 | * Note: this is probably too low level an operation for use in drivers. | |
765 | * Please consult with lkml before using this in your driver. | |
766 | */ | |
767 | void split_page(struct page *page, unsigned int order) | |
768 | { | |
769 | int i; | |
770 | ||
771 | BUG_ON(PageCompound(page)); | |
772 | BUG_ON(!page_count(page)); | |
7835e98b NP |
773 | for (i = 1; i < (1 << order); i++) |
774 | set_page_refcounted(page + i); | |
8dfcc9ba | 775 | } |
8dfcc9ba | 776 | |
1da177e4 LT |
777 | /* |
778 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But | |
779 | * we cheat by calling it from here, in the order > 0 path. Saves a branch | |
780 | * or two. | |
781 | */ | |
a74609fa NP |
782 | static struct page *buffered_rmqueue(struct zonelist *zonelist, |
783 | struct zone *zone, int order, gfp_t gfp_flags) | |
1da177e4 LT |
784 | { |
785 | unsigned long flags; | |
689bcebf | 786 | struct page *page; |
1da177e4 | 787 | int cold = !!(gfp_flags & __GFP_COLD); |
a74609fa | 788 | int cpu; |
1da177e4 | 789 | |
689bcebf | 790 | again: |
a74609fa | 791 | cpu = get_cpu(); |
48db57f8 | 792 | if (likely(order == 0)) { |
1da177e4 LT |
793 | struct per_cpu_pages *pcp; |
794 | ||
a74609fa | 795 | pcp = &zone_pcp(zone, cpu)->pcp[cold]; |
1da177e4 | 796 | local_irq_save(flags); |
a74609fa | 797 | if (!pcp->count) { |
1da177e4 LT |
798 | pcp->count += rmqueue_bulk(zone, 0, |
799 | pcp->batch, &pcp->list); | |
a74609fa NP |
800 | if (unlikely(!pcp->count)) |
801 | goto failed; | |
1da177e4 | 802 | } |
a74609fa NP |
803 | page = list_entry(pcp->list.next, struct page, lru); |
804 | list_del(&page->lru); | |
805 | pcp->count--; | |
7fb1d9fc | 806 | } else { |
1da177e4 LT |
807 | spin_lock_irqsave(&zone->lock, flags); |
808 | page = __rmqueue(zone, order); | |
a74609fa NP |
809 | spin_unlock(&zone->lock); |
810 | if (!page) | |
811 | goto failed; | |
1da177e4 LT |
812 | } |
813 | ||
a74609fa NP |
814 | __mod_page_state_zone(zone, pgalloc, 1 << order); |
815 | zone_statistics(zonelist, zone, cpu); | |
816 | local_irq_restore(flags); | |
817 | put_cpu(); | |
1da177e4 | 818 | |
a74609fa | 819 | BUG_ON(bad_range(zone, page)); |
17cf4406 | 820 | if (prep_new_page(page, order, gfp_flags)) |
a74609fa | 821 | goto again; |
1da177e4 | 822 | return page; |
a74609fa NP |
823 | |
824 | failed: | |
825 | local_irq_restore(flags); | |
826 | put_cpu(); | |
827 | return NULL; | |
1da177e4 LT |
828 | } |
829 | ||
7fb1d9fc | 830 | #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */ |
3148890b NP |
831 | #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */ |
832 | #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */ | |
833 | #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */ | |
834 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ | |
835 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ | |
836 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ | |
7fb1d9fc | 837 | |
1da177e4 LT |
838 | /* |
839 | * Return 1 if free pages are above 'mark'. This takes into account the order | |
840 | * of the allocation. | |
841 | */ | |
842 | int zone_watermark_ok(struct zone *z, int order, unsigned long mark, | |
7fb1d9fc | 843 | int classzone_idx, int alloc_flags) |
1da177e4 LT |
844 | { |
845 | /* free_pages my go negative - that's OK */ | |
846 | long min = mark, free_pages = z->free_pages - (1 << order) + 1; | |
847 | int o; | |
848 | ||
7fb1d9fc | 849 | if (alloc_flags & ALLOC_HIGH) |
1da177e4 | 850 | min -= min / 2; |
7fb1d9fc | 851 | if (alloc_flags & ALLOC_HARDER) |
1da177e4 LT |
852 | min -= min / 4; |
853 | ||
854 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) | |
855 | return 0; | |
856 | for (o = 0; o < order; o++) { | |
857 | /* At the next order, this order's pages become unavailable */ | |
858 | free_pages -= z->free_area[o].nr_free << o; | |
859 | ||
860 | /* Require fewer higher order pages to be free */ | |
861 | min >>= 1; | |
862 | ||
863 | if (free_pages <= min) | |
864 | return 0; | |
865 | } | |
866 | return 1; | |
867 | } | |
868 | ||
7fb1d9fc RS |
869 | /* |
870 | * get_page_from_freeliest goes through the zonelist trying to allocate | |
871 | * a page. | |
872 | */ | |
873 | static struct page * | |
874 | get_page_from_freelist(gfp_t gfp_mask, unsigned int order, | |
875 | struct zonelist *zonelist, int alloc_flags) | |
753ee728 | 876 | { |
7fb1d9fc RS |
877 | struct zone **z = zonelist->zones; |
878 | struct page *page = NULL; | |
879 | int classzone_idx = zone_idx(*z); | |
880 | ||
881 | /* | |
882 | * Go through the zonelist once, looking for a zone with enough free. | |
883 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. | |
884 | */ | |
885 | do { | |
886 | if ((alloc_flags & ALLOC_CPUSET) && | |
887 | !cpuset_zone_allowed(*z, gfp_mask)) | |
888 | continue; | |
889 | ||
890 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { | |
3148890b NP |
891 | unsigned long mark; |
892 | if (alloc_flags & ALLOC_WMARK_MIN) | |
893 | mark = (*z)->pages_min; | |
894 | else if (alloc_flags & ALLOC_WMARK_LOW) | |
895 | mark = (*z)->pages_low; | |
896 | else | |
897 | mark = (*z)->pages_high; | |
898 | if (!zone_watermark_ok(*z, order, mark, | |
7fb1d9fc | 899 | classzone_idx, alloc_flags)) |
9eeff239 CL |
900 | if (!zone_reclaim_mode || |
901 | !zone_reclaim(*z, gfp_mask, order)) | |
902 | continue; | |
7fb1d9fc RS |
903 | } |
904 | ||
a74609fa | 905 | page = buffered_rmqueue(zonelist, *z, order, gfp_mask); |
7fb1d9fc | 906 | if (page) { |
7fb1d9fc RS |
907 | break; |
908 | } | |
909 | } while (*(++z) != NULL); | |
910 | return page; | |
753ee728 MH |
911 | } |
912 | ||
1da177e4 LT |
913 | /* |
914 | * This is the 'heart' of the zoned buddy allocator. | |
915 | */ | |
916 | struct page * fastcall | |
dd0fc66f | 917 | __alloc_pages(gfp_t gfp_mask, unsigned int order, |
1da177e4 LT |
918 | struct zonelist *zonelist) |
919 | { | |
260b2367 | 920 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
7fb1d9fc | 921 | struct zone **z; |
1da177e4 LT |
922 | struct page *page; |
923 | struct reclaim_state reclaim_state; | |
924 | struct task_struct *p = current; | |
1da177e4 | 925 | int do_retry; |
7fb1d9fc | 926 | int alloc_flags; |
1da177e4 LT |
927 | int did_some_progress; |
928 | ||
929 | might_sleep_if(wait); | |
930 | ||
6b1de916 | 931 | restart: |
7fb1d9fc | 932 | z = zonelist->zones; /* the list of zones suitable for gfp_mask */ |
1da177e4 | 933 | |
7fb1d9fc | 934 | if (unlikely(*z == NULL)) { |
1da177e4 LT |
935 | /* Should this ever happen?? */ |
936 | return NULL; | |
937 | } | |
6b1de916 | 938 | |
7fb1d9fc | 939 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 940 | zonelist, ALLOC_WMARK_LOW|ALLOC_CPUSET); |
7fb1d9fc RS |
941 | if (page) |
942 | goto got_pg; | |
1da177e4 | 943 | |
6b1de916 | 944 | do { |
0b1303fc CL |
945 | if (cpuset_zone_allowed(*z, gfp_mask)) |
946 | wakeup_kswapd(*z, order); | |
6b1de916 | 947 | } while (*(++z)); |
1da177e4 | 948 | |
9bf2229f | 949 | /* |
7fb1d9fc RS |
950 | * OK, we're below the kswapd watermark and have kicked background |
951 | * reclaim. Now things get more complex, so set up alloc_flags according | |
952 | * to how we want to proceed. | |
953 | * | |
954 | * The caller may dip into page reserves a bit more if the caller | |
955 | * cannot run direct reclaim, or if the caller has realtime scheduling | |
4eac915d PJ |
956 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
957 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). | |
9bf2229f | 958 | */ |
3148890b | 959 | alloc_flags = ALLOC_WMARK_MIN; |
7fb1d9fc RS |
960 | if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait) |
961 | alloc_flags |= ALLOC_HARDER; | |
962 | if (gfp_mask & __GFP_HIGH) | |
963 | alloc_flags |= ALLOC_HIGH; | |
47f3a867 | 964 | alloc_flags |= ALLOC_CPUSET; |
1da177e4 LT |
965 | |
966 | /* | |
967 | * Go through the zonelist again. Let __GFP_HIGH and allocations | |
7fb1d9fc | 968 | * coming from realtime tasks go deeper into reserves. |
1da177e4 LT |
969 | * |
970 | * This is the last chance, in general, before the goto nopage. | |
971 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. | |
9bf2229f | 972 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
1da177e4 | 973 | */ |
7fb1d9fc RS |
974 | page = get_page_from_freelist(gfp_mask, order, zonelist, alloc_flags); |
975 | if (page) | |
976 | goto got_pg; | |
1da177e4 LT |
977 | |
978 | /* This allocation should allow future memory freeing. */ | |
b84a35be NP |
979 | |
980 | if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE))) | |
981 | && !in_interrupt()) { | |
982 | if (!(gfp_mask & __GFP_NOMEMALLOC)) { | |
885036d3 | 983 | nofail_alloc: |
b84a35be | 984 | /* go through the zonelist yet again, ignoring mins */ |
7fb1d9fc | 985 | page = get_page_from_freelist(gfp_mask, order, |
47f3a867 | 986 | zonelist, ALLOC_NO_WATERMARKS); |
7fb1d9fc RS |
987 | if (page) |
988 | goto got_pg; | |
885036d3 KK |
989 | if (gfp_mask & __GFP_NOFAIL) { |
990 | blk_congestion_wait(WRITE, HZ/50); | |
991 | goto nofail_alloc; | |
992 | } | |
1da177e4 LT |
993 | } |
994 | goto nopage; | |
995 | } | |
996 | ||
997 | /* Atomic allocations - we can't balance anything */ | |
998 | if (!wait) | |
999 | goto nopage; | |
1000 | ||
1001 | rebalance: | |
1002 | cond_resched(); | |
1003 | ||
1004 | /* We now go into synchronous reclaim */ | |
3e0d98b9 | 1005 | cpuset_memory_pressure_bump(); |
1da177e4 LT |
1006 | p->flags |= PF_MEMALLOC; |
1007 | reclaim_state.reclaimed_slab = 0; | |
1008 | p->reclaim_state = &reclaim_state; | |
1009 | ||
7fb1d9fc | 1010 | did_some_progress = try_to_free_pages(zonelist->zones, gfp_mask); |
1da177e4 LT |
1011 | |
1012 | p->reclaim_state = NULL; | |
1013 | p->flags &= ~PF_MEMALLOC; | |
1014 | ||
1015 | cond_resched(); | |
1016 | ||
1017 | if (likely(did_some_progress)) { | |
7fb1d9fc RS |
1018 | page = get_page_from_freelist(gfp_mask, order, |
1019 | zonelist, alloc_flags); | |
1020 | if (page) | |
1021 | goto got_pg; | |
1da177e4 LT |
1022 | } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
1023 | /* | |
1024 | * Go through the zonelist yet one more time, keep | |
1025 | * very high watermark here, this is only to catch | |
1026 | * a parallel oom killing, we must fail if we're still | |
1027 | * under heavy pressure. | |
1028 | */ | |
7fb1d9fc | 1029 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, order, |
3148890b | 1030 | zonelist, ALLOC_WMARK_HIGH|ALLOC_CPUSET); |
7fb1d9fc RS |
1031 | if (page) |
1032 | goto got_pg; | |
1da177e4 | 1033 | |
9b0f8b04 | 1034 | out_of_memory(zonelist, gfp_mask, order); |
1da177e4 LT |
1035 | goto restart; |
1036 | } | |
1037 | ||
1038 | /* | |
1039 | * Don't let big-order allocations loop unless the caller explicitly | |
1040 | * requests that. Wait for some write requests to complete then retry. | |
1041 | * | |
1042 | * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order | |
1043 | * <= 3, but that may not be true in other implementations. | |
1044 | */ | |
1045 | do_retry = 0; | |
1046 | if (!(gfp_mask & __GFP_NORETRY)) { | |
1047 | if ((order <= 3) || (gfp_mask & __GFP_REPEAT)) | |
1048 | do_retry = 1; | |
1049 | if (gfp_mask & __GFP_NOFAIL) | |
1050 | do_retry = 1; | |
1051 | } | |
1052 | if (do_retry) { | |
1053 | blk_congestion_wait(WRITE, HZ/50); | |
1054 | goto rebalance; | |
1055 | } | |
1056 | ||
1057 | nopage: | |
1058 | if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) { | |
1059 | printk(KERN_WARNING "%s: page allocation failure." | |
1060 | " order:%d, mode:0x%x\n", | |
1061 | p->comm, order, gfp_mask); | |
1062 | dump_stack(); | |
578c2fd6 | 1063 | show_mem(); |
1da177e4 | 1064 | } |
1da177e4 | 1065 | got_pg: |
1da177e4 LT |
1066 | return page; |
1067 | } | |
1068 | ||
1069 | EXPORT_SYMBOL(__alloc_pages); | |
1070 | ||
1071 | /* | |
1072 | * Common helper functions. | |
1073 | */ | |
dd0fc66f | 1074 | fastcall unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
1da177e4 LT |
1075 | { |
1076 | struct page * page; | |
1077 | page = alloc_pages(gfp_mask, order); | |
1078 | if (!page) | |
1079 | return 0; | |
1080 | return (unsigned long) page_address(page); | |
1081 | } | |
1082 | ||
1083 | EXPORT_SYMBOL(__get_free_pages); | |
1084 | ||
dd0fc66f | 1085 | fastcall unsigned long get_zeroed_page(gfp_t gfp_mask) |
1da177e4 LT |
1086 | { |
1087 | struct page * page; | |
1088 | ||
1089 | /* | |
1090 | * get_zeroed_page() returns a 32-bit address, which cannot represent | |
1091 | * a highmem page | |
1092 | */ | |
260b2367 | 1093 | BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
1da177e4 LT |
1094 | |
1095 | page = alloc_pages(gfp_mask | __GFP_ZERO, 0); | |
1096 | if (page) | |
1097 | return (unsigned long) page_address(page); | |
1098 | return 0; | |
1099 | } | |
1100 | ||
1101 | EXPORT_SYMBOL(get_zeroed_page); | |
1102 | ||
1103 | void __pagevec_free(struct pagevec *pvec) | |
1104 | { | |
1105 | int i = pagevec_count(pvec); | |
1106 | ||
1107 | while (--i >= 0) | |
1108 | free_hot_cold_page(pvec->pages[i], pvec->cold); | |
1109 | } | |
1110 | ||
1111 | fastcall void __free_pages(struct page *page, unsigned int order) | |
1112 | { | |
b5810039 | 1113 | if (put_page_testzero(page)) { |
1da177e4 LT |
1114 | if (order == 0) |
1115 | free_hot_page(page); | |
1116 | else | |
1117 | __free_pages_ok(page, order); | |
1118 | } | |
1119 | } | |
1120 | ||
1121 | EXPORT_SYMBOL(__free_pages); | |
1122 | ||
1123 | fastcall void free_pages(unsigned long addr, unsigned int order) | |
1124 | { | |
1125 | if (addr != 0) { | |
1126 | BUG_ON(!virt_addr_valid((void *)addr)); | |
1127 | __free_pages(virt_to_page((void *)addr), order); | |
1128 | } | |
1129 | } | |
1130 | ||
1131 | EXPORT_SYMBOL(free_pages); | |
1132 | ||
1133 | /* | |
1134 | * Total amount of free (allocatable) RAM: | |
1135 | */ | |
1136 | unsigned int nr_free_pages(void) | |
1137 | { | |
1138 | unsigned int sum = 0; | |
1139 | struct zone *zone; | |
1140 | ||
1141 | for_each_zone(zone) | |
1142 | sum += zone->free_pages; | |
1143 | ||
1144 | return sum; | |
1145 | } | |
1146 | ||
1147 | EXPORT_SYMBOL(nr_free_pages); | |
1148 | ||
1149 | #ifdef CONFIG_NUMA | |
1150 | unsigned int nr_free_pages_pgdat(pg_data_t *pgdat) | |
1151 | { | |
1152 | unsigned int i, sum = 0; | |
1153 | ||
1154 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1155 | sum += pgdat->node_zones[i].free_pages; | |
1156 | ||
1157 | return sum; | |
1158 | } | |
1159 | #endif | |
1160 | ||
1161 | static unsigned int nr_free_zone_pages(int offset) | |
1162 | { | |
e310fd43 MB |
1163 | /* Just pick one node, since fallback list is circular */ |
1164 | pg_data_t *pgdat = NODE_DATA(numa_node_id()); | |
1da177e4 LT |
1165 | unsigned int sum = 0; |
1166 | ||
e310fd43 MB |
1167 | struct zonelist *zonelist = pgdat->node_zonelists + offset; |
1168 | struct zone **zonep = zonelist->zones; | |
1169 | struct zone *zone; | |
1da177e4 | 1170 | |
e310fd43 MB |
1171 | for (zone = *zonep++; zone; zone = *zonep++) { |
1172 | unsigned long size = zone->present_pages; | |
1173 | unsigned long high = zone->pages_high; | |
1174 | if (size > high) | |
1175 | sum += size - high; | |
1da177e4 LT |
1176 | } |
1177 | ||
1178 | return sum; | |
1179 | } | |
1180 | ||
1181 | /* | |
1182 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL | |
1183 | */ | |
1184 | unsigned int nr_free_buffer_pages(void) | |
1185 | { | |
af4ca457 | 1186 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
1da177e4 LT |
1187 | } |
1188 | ||
1189 | /* | |
1190 | * Amount of free RAM allocatable within all zones | |
1191 | */ | |
1192 | unsigned int nr_free_pagecache_pages(void) | |
1193 | { | |
af4ca457 | 1194 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER)); |
1da177e4 LT |
1195 | } |
1196 | ||
1197 | #ifdef CONFIG_HIGHMEM | |
1198 | unsigned int nr_free_highpages (void) | |
1199 | { | |
1200 | pg_data_t *pgdat; | |
1201 | unsigned int pages = 0; | |
1202 | ||
ec936fc5 | 1203 | for_each_online_pgdat(pgdat) |
1da177e4 LT |
1204 | pages += pgdat->node_zones[ZONE_HIGHMEM].free_pages; |
1205 | ||
1206 | return pages; | |
1207 | } | |
1208 | #endif | |
1209 | ||
1210 | #ifdef CONFIG_NUMA | |
1211 | static void show_node(struct zone *zone) | |
1212 | { | |
1213 | printk("Node %d ", zone->zone_pgdat->node_id); | |
1214 | } | |
1215 | #else | |
1216 | #define show_node(zone) do { } while (0) | |
1217 | #endif | |
1218 | ||
1219 | /* | |
1220 | * Accumulate the page_state information across all CPUs. | |
1221 | * The result is unavoidably approximate - it can change | |
1222 | * during and after execution of this function. | |
1223 | */ | |
1224 | static DEFINE_PER_CPU(struct page_state, page_states) = {0}; | |
1225 | ||
1226 | atomic_t nr_pagecache = ATOMIC_INIT(0); | |
1227 | EXPORT_SYMBOL(nr_pagecache); | |
1228 | #ifdef CONFIG_SMP | |
1229 | DEFINE_PER_CPU(long, nr_pagecache_local) = 0; | |
1230 | #endif | |
1231 | ||
a86b1f53 | 1232 | static void __get_page_state(struct page_state *ret, int nr, cpumask_t *cpumask) |
1da177e4 | 1233 | { |
b40607fc | 1234 | unsigned cpu; |
1da177e4 | 1235 | |
88a2a4ac | 1236 | memset(ret, 0, nr * sizeof(unsigned long)); |
84c2008a | 1237 | cpus_and(*cpumask, *cpumask, cpu_online_map); |
1da177e4 | 1238 | |
b40607fc AM |
1239 | for_each_cpu_mask(cpu, *cpumask) { |
1240 | unsigned long *in; | |
1241 | unsigned long *out; | |
1242 | unsigned off; | |
1243 | unsigned next_cpu; | |
88a2a4ac | 1244 | |
1da177e4 LT |
1245 | in = (unsigned long *)&per_cpu(page_states, cpu); |
1246 | ||
b40607fc AM |
1247 | next_cpu = next_cpu(cpu, *cpumask); |
1248 | if (likely(next_cpu < NR_CPUS)) | |
1249 | prefetch(&per_cpu(page_states, next_cpu)); | |
1da177e4 LT |
1250 | |
1251 | out = (unsigned long *)ret; | |
1252 | for (off = 0; off < nr; off++) | |
1253 | *out++ += *in++; | |
1254 | } | |
1255 | } | |
1256 | ||
c07e02db MH |
1257 | void get_page_state_node(struct page_state *ret, int node) |
1258 | { | |
1259 | int nr; | |
1260 | cpumask_t mask = node_to_cpumask(node); | |
1261 | ||
1262 | nr = offsetof(struct page_state, GET_PAGE_STATE_LAST); | |
1263 | nr /= sizeof(unsigned long); | |
1264 | ||
1265 | __get_page_state(ret, nr+1, &mask); | |
1266 | } | |
1267 | ||
1da177e4 LT |
1268 | void get_page_state(struct page_state *ret) |
1269 | { | |
1270 | int nr; | |
c07e02db | 1271 | cpumask_t mask = CPU_MASK_ALL; |
1da177e4 LT |
1272 | |
1273 | nr = offsetof(struct page_state, GET_PAGE_STATE_LAST); | |
1274 | nr /= sizeof(unsigned long); | |
1275 | ||
c07e02db | 1276 | __get_page_state(ret, nr + 1, &mask); |
1da177e4 LT |
1277 | } |
1278 | ||
1279 | void get_full_page_state(struct page_state *ret) | |
1280 | { | |
c07e02db MH |
1281 | cpumask_t mask = CPU_MASK_ALL; |
1282 | ||
1283 | __get_page_state(ret, sizeof(*ret) / sizeof(unsigned long), &mask); | |
1da177e4 LT |
1284 | } |
1285 | ||
a74609fa | 1286 | unsigned long read_page_state_offset(unsigned long offset) |
1da177e4 LT |
1287 | { |
1288 | unsigned long ret = 0; | |
1289 | int cpu; | |
1290 | ||
84c2008a | 1291 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1292 | unsigned long in; |
1293 | ||
1294 | in = (unsigned long)&per_cpu(page_states, cpu) + offset; | |
1295 | ret += *((unsigned long *)in); | |
1296 | } | |
1297 | return ret; | |
1298 | } | |
1299 | ||
a74609fa NP |
1300 | void __mod_page_state_offset(unsigned long offset, unsigned long delta) |
1301 | { | |
1302 | void *ptr; | |
1303 | ||
1304 | ptr = &__get_cpu_var(page_states); | |
1305 | *(unsigned long *)(ptr + offset) += delta; | |
1306 | } | |
1307 | EXPORT_SYMBOL(__mod_page_state_offset); | |
1308 | ||
1309 | void mod_page_state_offset(unsigned long offset, unsigned long delta) | |
1da177e4 LT |
1310 | { |
1311 | unsigned long flags; | |
a74609fa | 1312 | void *ptr; |
1da177e4 LT |
1313 | |
1314 | local_irq_save(flags); | |
1315 | ptr = &__get_cpu_var(page_states); | |
a74609fa | 1316 | *(unsigned long *)(ptr + offset) += delta; |
1da177e4 LT |
1317 | local_irq_restore(flags); |
1318 | } | |
a74609fa | 1319 | EXPORT_SYMBOL(mod_page_state_offset); |
1da177e4 LT |
1320 | |
1321 | void __get_zone_counts(unsigned long *active, unsigned long *inactive, | |
1322 | unsigned long *free, struct pglist_data *pgdat) | |
1323 | { | |
1324 | struct zone *zones = pgdat->node_zones; | |
1325 | int i; | |
1326 | ||
1327 | *active = 0; | |
1328 | *inactive = 0; | |
1329 | *free = 0; | |
1330 | for (i = 0; i < MAX_NR_ZONES; i++) { | |
1331 | *active += zones[i].nr_active; | |
1332 | *inactive += zones[i].nr_inactive; | |
1333 | *free += zones[i].free_pages; | |
1334 | } | |
1335 | } | |
1336 | ||
1337 | void get_zone_counts(unsigned long *active, | |
1338 | unsigned long *inactive, unsigned long *free) | |
1339 | { | |
1340 | struct pglist_data *pgdat; | |
1341 | ||
1342 | *active = 0; | |
1343 | *inactive = 0; | |
1344 | *free = 0; | |
ec936fc5 | 1345 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
1346 | unsigned long l, m, n; |
1347 | __get_zone_counts(&l, &m, &n, pgdat); | |
1348 | *active += l; | |
1349 | *inactive += m; | |
1350 | *free += n; | |
1351 | } | |
1352 | } | |
1353 | ||
1354 | void si_meminfo(struct sysinfo *val) | |
1355 | { | |
1356 | val->totalram = totalram_pages; | |
1357 | val->sharedram = 0; | |
1358 | val->freeram = nr_free_pages(); | |
1359 | val->bufferram = nr_blockdev_pages(); | |
1360 | #ifdef CONFIG_HIGHMEM | |
1361 | val->totalhigh = totalhigh_pages; | |
1362 | val->freehigh = nr_free_highpages(); | |
1363 | #else | |
1364 | val->totalhigh = 0; | |
1365 | val->freehigh = 0; | |
1366 | #endif | |
1367 | val->mem_unit = PAGE_SIZE; | |
1368 | } | |
1369 | ||
1370 | EXPORT_SYMBOL(si_meminfo); | |
1371 | ||
1372 | #ifdef CONFIG_NUMA | |
1373 | void si_meminfo_node(struct sysinfo *val, int nid) | |
1374 | { | |
1375 | pg_data_t *pgdat = NODE_DATA(nid); | |
1376 | ||
1377 | val->totalram = pgdat->node_present_pages; | |
1378 | val->freeram = nr_free_pages_pgdat(pgdat); | |
1379 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; | |
1380 | val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages; | |
1381 | val->mem_unit = PAGE_SIZE; | |
1382 | } | |
1383 | #endif | |
1384 | ||
1385 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
1386 | ||
1387 | /* | |
1388 | * Show free area list (used inside shift_scroll-lock stuff) | |
1389 | * We also calculate the percentage fragmentation. We do this by counting the | |
1390 | * memory on each free list with the exception of the first item on the list. | |
1391 | */ | |
1392 | void show_free_areas(void) | |
1393 | { | |
1394 | struct page_state ps; | |
1395 | int cpu, temperature; | |
1396 | unsigned long active; | |
1397 | unsigned long inactive; | |
1398 | unsigned long free; | |
1399 | struct zone *zone; | |
1400 | ||
1401 | for_each_zone(zone) { | |
1402 | show_node(zone); | |
1403 | printk("%s per-cpu:", zone->name); | |
1404 | ||
f3fe6512 | 1405 | if (!populated_zone(zone)) { |
1da177e4 LT |
1406 | printk(" empty\n"); |
1407 | continue; | |
1408 | } else | |
1409 | printk("\n"); | |
1410 | ||
6b482c67 | 1411 | for_each_online_cpu(cpu) { |
1da177e4 LT |
1412 | struct per_cpu_pageset *pageset; |
1413 | ||
e7c8d5c9 | 1414 | pageset = zone_pcp(zone, cpu); |
1da177e4 LT |
1415 | |
1416 | for (temperature = 0; temperature < 2; temperature++) | |
2d92c5c9 | 1417 | printk("cpu %d %s: high %d, batch %d used:%d\n", |
1da177e4 LT |
1418 | cpu, |
1419 | temperature ? "cold" : "hot", | |
1da177e4 | 1420 | pageset->pcp[temperature].high, |
4ae7c039 CL |
1421 | pageset->pcp[temperature].batch, |
1422 | pageset->pcp[temperature].count); | |
1da177e4 LT |
1423 | } |
1424 | } | |
1425 | ||
1426 | get_page_state(&ps); | |
1427 | get_zone_counts(&active, &inactive, &free); | |
1428 | ||
c0d62219 | 1429 | printk("Free pages: %11ukB (%ukB HighMem)\n", |
1da177e4 LT |
1430 | K(nr_free_pages()), |
1431 | K(nr_free_highpages())); | |
1432 | ||
1433 | printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu " | |
1434 | "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n", | |
1435 | active, | |
1436 | inactive, | |
1437 | ps.nr_dirty, | |
1438 | ps.nr_writeback, | |
1439 | ps.nr_unstable, | |
1440 | nr_free_pages(), | |
1441 | ps.nr_slab, | |
1442 | ps.nr_mapped, | |
1443 | ps.nr_page_table_pages); | |
1444 | ||
1445 | for_each_zone(zone) { | |
1446 | int i; | |
1447 | ||
1448 | show_node(zone); | |
1449 | printk("%s" | |
1450 | " free:%lukB" | |
1451 | " min:%lukB" | |
1452 | " low:%lukB" | |
1453 | " high:%lukB" | |
1454 | " active:%lukB" | |
1455 | " inactive:%lukB" | |
1456 | " present:%lukB" | |
1457 | " pages_scanned:%lu" | |
1458 | " all_unreclaimable? %s" | |
1459 | "\n", | |
1460 | zone->name, | |
1461 | K(zone->free_pages), | |
1462 | K(zone->pages_min), | |
1463 | K(zone->pages_low), | |
1464 | K(zone->pages_high), | |
1465 | K(zone->nr_active), | |
1466 | K(zone->nr_inactive), | |
1467 | K(zone->present_pages), | |
1468 | zone->pages_scanned, | |
1469 | (zone->all_unreclaimable ? "yes" : "no") | |
1470 | ); | |
1471 | printk("lowmem_reserve[]:"); | |
1472 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1473 | printk(" %lu", zone->lowmem_reserve[i]); | |
1474 | printk("\n"); | |
1475 | } | |
1476 | ||
1477 | for_each_zone(zone) { | |
1478 | unsigned long nr, flags, order, total = 0; | |
1479 | ||
1480 | show_node(zone); | |
1481 | printk("%s: ", zone->name); | |
f3fe6512 | 1482 | if (!populated_zone(zone)) { |
1da177e4 LT |
1483 | printk("empty\n"); |
1484 | continue; | |
1485 | } | |
1486 | ||
1487 | spin_lock_irqsave(&zone->lock, flags); | |
1488 | for (order = 0; order < MAX_ORDER; order++) { | |
1489 | nr = zone->free_area[order].nr_free; | |
1490 | total += nr << order; | |
1491 | printk("%lu*%lukB ", nr, K(1UL) << order); | |
1492 | } | |
1493 | spin_unlock_irqrestore(&zone->lock, flags); | |
1494 | printk("= %lukB\n", K(total)); | |
1495 | } | |
1496 | ||
1497 | show_swap_cache_info(); | |
1498 | } | |
1499 | ||
1500 | /* | |
1501 | * Builds allocation fallback zone lists. | |
1a93205b CL |
1502 | * |
1503 | * Add all populated zones of a node to the zonelist. | |
1da177e4 | 1504 | */ |
1a93205b | 1505 | static int __init build_zonelists_node(pg_data_t *pgdat, |
070f8032 | 1506 | struct zonelist *zonelist, int nr_zones, int zone_type) |
1da177e4 | 1507 | { |
1a93205b CL |
1508 | struct zone *zone; |
1509 | ||
070f8032 | 1510 | BUG_ON(zone_type > ZONE_HIGHMEM); |
02a68a5e CL |
1511 | |
1512 | do { | |
070f8032 | 1513 | zone = pgdat->node_zones + zone_type; |
1a93205b | 1514 | if (populated_zone(zone)) { |
1da177e4 | 1515 | #ifndef CONFIG_HIGHMEM |
070f8032 | 1516 | BUG_ON(zone_type > ZONE_NORMAL); |
1da177e4 | 1517 | #endif |
070f8032 CL |
1518 | zonelist->zones[nr_zones++] = zone; |
1519 | check_highest_zone(zone_type); | |
1da177e4 | 1520 | } |
070f8032 | 1521 | zone_type--; |
02a68a5e | 1522 | |
070f8032 CL |
1523 | } while (zone_type >= 0); |
1524 | return nr_zones; | |
1da177e4 LT |
1525 | } |
1526 | ||
260b2367 AV |
1527 | static inline int highest_zone(int zone_bits) |
1528 | { | |
1529 | int res = ZONE_NORMAL; | |
1530 | if (zone_bits & (__force int)__GFP_HIGHMEM) | |
1531 | res = ZONE_HIGHMEM; | |
a2f1b424 AK |
1532 | if (zone_bits & (__force int)__GFP_DMA32) |
1533 | res = ZONE_DMA32; | |
260b2367 AV |
1534 | if (zone_bits & (__force int)__GFP_DMA) |
1535 | res = ZONE_DMA; | |
1536 | return res; | |
1537 | } | |
1538 | ||
1da177e4 LT |
1539 | #ifdef CONFIG_NUMA |
1540 | #define MAX_NODE_LOAD (num_online_nodes()) | |
1541 | static int __initdata node_load[MAX_NUMNODES]; | |
1542 | /** | |
4dc3b16b | 1543 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
1da177e4 LT |
1544 | * @node: node whose fallback list we're appending |
1545 | * @used_node_mask: nodemask_t of already used nodes | |
1546 | * | |
1547 | * We use a number of factors to determine which is the next node that should | |
1548 | * appear on a given node's fallback list. The node should not have appeared | |
1549 | * already in @node's fallback list, and it should be the next closest node | |
1550 | * according to the distance array (which contains arbitrary distance values | |
1551 | * from each node to each node in the system), and should also prefer nodes | |
1552 | * with no CPUs, since presumably they'll have very little allocation pressure | |
1553 | * on them otherwise. | |
1554 | * It returns -1 if no node is found. | |
1555 | */ | |
1556 | static int __init find_next_best_node(int node, nodemask_t *used_node_mask) | |
1557 | { | |
4cf808eb | 1558 | int n, val; |
1da177e4 LT |
1559 | int min_val = INT_MAX; |
1560 | int best_node = -1; | |
1561 | ||
4cf808eb LT |
1562 | /* Use the local node if we haven't already */ |
1563 | if (!node_isset(node, *used_node_mask)) { | |
1564 | node_set(node, *used_node_mask); | |
1565 | return node; | |
1566 | } | |
1da177e4 | 1567 | |
4cf808eb LT |
1568 | for_each_online_node(n) { |
1569 | cpumask_t tmp; | |
1da177e4 LT |
1570 | |
1571 | /* Don't want a node to appear more than once */ | |
1572 | if (node_isset(n, *used_node_mask)) | |
1573 | continue; | |
1574 | ||
1da177e4 LT |
1575 | /* Use the distance array to find the distance */ |
1576 | val = node_distance(node, n); | |
1577 | ||
4cf808eb LT |
1578 | /* Penalize nodes under us ("prefer the next node") */ |
1579 | val += (n < node); | |
1580 | ||
1da177e4 LT |
1581 | /* Give preference to headless and unused nodes */ |
1582 | tmp = node_to_cpumask(n); | |
1583 | if (!cpus_empty(tmp)) | |
1584 | val += PENALTY_FOR_NODE_WITH_CPUS; | |
1585 | ||
1586 | /* Slight preference for less loaded node */ | |
1587 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); | |
1588 | val += node_load[n]; | |
1589 | ||
1590 | if (val < min_val) { | |
1591 | min_val = val; | |
1592 | best_node = n; | |
1593 | } | |
1594 | } | |
1595 | ||
1596 | if (best_node >= 0) | |
1597 | node_set(best_node, *used_node_mask); | |
1598 | ||
1599 | return best_node; | |
1600 | } | |
1601 | ||
1602 | static void __init build_zonelists(pg_data_t *pgdat) | |
1603 | { | |
1604 | int i, j, k, node, local_node; | |
1605 | int prev_node, load; | |
1606 | struct zonelist *zonelist; | |
1607 | nodemask_t used_mask; | |
1608 | ||
1609 | /* initialize zonelists */ | |
1610 | for (i = 0; i < GFP_ZONETYPES; i++) { | |
1611 | zonelist = pgdat->node_zonelists + i; | |
1612 | zonelist->zones[0] = NULL; | |
1613 | } | |
1614 | ||
1615 | /* NUMA-aware ordering of nodes */ | |
1616 | local_node = pgdat->node_id; | |
1617 | load = num_online_nodes(); | |
1618 | prev_node = local_node; | |
1619 | nodes_clear(used_mask); | |
1620 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { | |
9eeff239 CL |
1621 | int distance = node_distance(local_node, node); |
1622 | ||
1623 | /* | |
1624 | * If another node is sufficiently far away then it is better | |
1625 | * to reclaim pages in a zone before going off node. | |
1626 | */ | |
1627 | if (distance > RECLAIM_DISTANCE) | |
1628 | zone_reclaim_mode = 1; | |
1629 | ||
1da177e4 LT |
1630 | /* |
1631 | * We don't want to pressure a particular node. | |
1632 | * So adding penalty to the first node in same | |
1633 | * distance group to make it round-robin. | |
1634 | */ | |
9eeff239 CL |
1635 | |
1636 | if (distance != node_distance(local_node, prev_node)) | |
1da177e4 LT |
1637 | node_load[node] += load; |
1638 | prev_node = node; | |
1639 | load--; | |
1640 | for (i = 0; i < GFP_ZONETYPES; i++) { | |
1641 | zonelist = pgdat->node_zonelists + i; | |
1642 | for (j = 0; zonelist->zones[j] != NULL; j++); | |
1643 | ||
260b2367 | 1644 | k = highest_zone(i); |
1da177e4 LT |
1645 | |
1646 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, k); | |
1647 | zonelist->zones[j] = NULL; | |
1648 | } | |
1649 | } | |
1650 | } | |
1651 | ||
1652 | #else /* CONFIG_NUMA */ | |
1653 | ||
1654 | static void __init build_zonelists(pg_data_t *pgdat) | |
1655 | { | |
1656 | int i, j, k, node, local_node; | |
1657 | ||
1658 | local_node = pgdat->node_id; | |
1659 | for (i = 0; i < GFP_ZONETYPES; i++) { | |
1660 | struct zonelist *zonelist; | |
1661 | ||
1662 | zonelist = pgdat->node_zonelists + i; | |
1663 | ||
1664 | j = 0; | |
260b2367 | 1665 | k = highest_zone(i); |
1da177e4 LT |
1666 | j = build_zonelists_node(pgdat, zonelist, j, k); |
1667 | /* | |
1668 | * Now we build the zonelist so that it contains the zones | |
1669 | * of all the other nodes. | |
1670 | * We don't want to pressure a particular node, so when | |
1671 | * building the zones for node N, we make sure that the | |
1672 | * zones coming right after the local ones are those from | |
1673 | * node N+1 (modulo N) | |
1674 | */ | |
1675 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { | |
1676 | if (!node_online(node)) | |
1677 | continue; | |
1678 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, k); | |
1679 | } | |
1680 | for (node = 0; node < local_node; node++) { | |
1681 | if (!node_online(node)) | |
1682 | continue; | |
1683 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, k); | |
1684 | } | |
1685 | ||
1686 | zonelist->zones[j] = NULL; | |
1687 | } | |
1688 | } | |
1689 | ||
1690 | #endif /* CONFIG_NUMA */ | |
1691 | ||
1692 | void __init build_all_zonelists(void) | |
1693 | { | |
1694 | int i; | |
1695 | ||
1696 | for_each_online_node(i) | |
1697 | build_zonelists(NODE_DATA(i)); | |
1698 | printk("Built %i zonelists\n", num_online_nodes()); | |
1699 | cpuset_init_current_mems_allowed(); | |
1700 | } | |
1701 | ||
1702 | /* | |
1703 | * Helper functions to size the waitqueue hash table. | |
1704 | * Essentially these want to choose hash table sizes sufficiently | |
1705 | * large so that collisions trying to wait on pages are rare. | |
1706 | * But in fact, the number of active page waitqueues on typical | |
1707 | * systems is ridiculously low, less than 200. So this is even | |
1708 | * conservative, even though it seems large. | |
1709 | * | |
1710 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to | |
1711 | * waitqueues, i.e. the size of the waitq table given the number of pages. | |
1712 | */ | |
1713 | #define PAGES_PER_WAITQUEUE 256 | |
1714 | ||
1715 | static inline unsigned long wait_table_size(unsigned long pages) | |
1716 | { | |
1717 | unsigned long size = 1; | |
1718 | ||
1719 | pages /= PAGES_PER_WAITQUEUE; | |
1720 | ||
1721 | while (size < pages) | |
1722 | size <<= 1; | |
1723 | ||
1724 | /* | |
1725 | * Once we have dozens or even hundreds of threads sleeping | |
1726 | * on IO we've got bigger problems than wait queue collision. | |
1727 | * Limit the size of the wait table to a reasonable size. | |
1728 | */ | |
1729 | size = min(size, 4096UL); | |
1730 | ||
1731 | return max(size, 4UL); | |
1732 | } | |
1733 | ||
1734 | /* | |
1735 | * This is an integer logarithm so that shifts can be used later | |
1736 | * to extract the more random high bits from the multiplicative | |
1737 | * hash function before the remainder is taken. | |
1738 | */ | |
1739 | static inline unsigned long wait_table_bits(unsigned long size) | |
1740 | { | |
1741 | return ffz(~size); | |
1742 | } | |
1743 | ||
1744 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) | |
1745 | ||
1746 | static void __init calculate_zone_totalpages(struct pglist_data *pgdat, | |
1747 | unsigned long *zones_size, unsigned long *zholes_size) | |
1748 | { | |
1749 | unsigned long realtotalpages, totalpages = 0; | |
1750 | int i; | |
1751 | ||
1752 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1753 | totalpages += zones_size[i]; | |
1754 | pgdat->node_spanned_pages = totalpages; | |
1755 | ||
1756 | realtotalpages = totalpages; | |
1757 | if (zholes_size) | |
1758 | for (i = 0; i < MAX_NR_ZONES; i++) | |
1759 | realtotalpages -= zholes_size[i]; | |
1760 | pgdat->node_present_pages = realtotalpages; | |
1761 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages); | |
1762 | } | |
1763 | ||
1764 | ||
1765 | /* | |
1766 | * Initially all pages are reserved - free ones are freed | |
1767 | * up by free_all_bootmem() once the early boot process is | |
1768 | * done. Non-atomic initialization, single-pass. | |
1769 | */ | |
c09b4240 | 1770 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
1da177e4 LT |
1771 | unsigned long start_pfn) |
1772 | { | |
1da177e4 | 1773 | struct page *page; |
29751f69 AW |
1774 | unsigned long end_pfn = start_pfn + size; |
1775 | unsigned long pfn; | |
1da177e4 | 1776 | |
cbe8dd4a | 1777 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
d41dee36 AW |
1778 | if (!early_pfn_valid(pfn)) |
1779 | continue; | |
1780 | page = pfn_to_page(pfn); | |
1781 | set_page_links(page, zone, nid, pfn); | |
7835e98b | 1782 | init_page_count(page); |
1da177e4 LT |
1783 | reset_page_mapcount(page); |
1784 | SetPageReserved(page); | |
1785 | INIT_LIST_HEAD(&page->lru); | |
1786 | #ifdef WANT_PAGE_VIRTUAL | |
1787 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ | |
1788 | if (!is_highmem_idx(zone)) | |
3212c6be | 1789 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
1da177e4 | 1790 | #endif |
1da177e4 LT |
1791 | } |
1792 | } | |
1793 | ||
1794 | void zone_init_free_lists(struct pglist_data *pgdat, struct zone *zone, | |
1795 | unsigned long size) | |
1796 | { | |
1797 | int order; | |
1798 | for (order = 0; order < MAX_ORDER ; order++) { | |
1799 | INIT_LIST_HEAD(&zone->free_area[order].free_list); | |
1800 | zone->free_area[order].nr_free = 0; | |
1801 | } | |
1802 | } | |
1803 | ||
d41dee36 AW |
1804 | #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr) |
1805 | void zonetable_add(struct zone *zone, int nid, int zid, unsigned long pfn, | |
1806 | unsigned long size) | |
1807 | { | |
1808 | unsigned long snum = pfn_to_section_nr(pfn); | |
1809 | unsigned long end = pfn_to_section_nr(pfn + size); | |
1810 | ||
1811 | if (FLAGS_HAS_NODE) | |
1812 | zone_table[ZONETABLE_INDEX(nid, zid)] = zone; | |
1813 | else | |
1814 | for (; snum <= end; snum++) | |
1815 | zone_table[ZONETABLE_INDEX(snum, zid)] = zone; | |
1816 | } | |
1817 | ||
1da177e4 LT |
1818 | #ifndef __HAVE_ARCH_MEMMAP_INIT |
1819 | #define memmap_init(size, nid, zone, start_pfn) \ | |
1820 | memmap_init_zone((size), (nid), (zone), (start_pfn)) | |
1821 | #endif | |
1822 | ||
6292d9aa | 1823 | static int __cpuinit zone_batchsize(struct zone *zone) |
e7c8d5c9 CL |
1824 | { |
1825 | int batch; | |
1826 | ||
1827 | /* | |
1828 | * The per-cpu-pages pools are set to around 1000th of the | |
ba56e91c | 1829 | * size of the zone. But no more than 1/2 of a meg. |
e7c8d5c9 CL |
1830 | * |
1831 | * OK, so we don't know how big the cache is. So guess. | |
1832 | */ | |
1833 | batch = zone->present_pages / 1024; | |
ba56e91c SR |
1834 | if (batch * PAGE_SIZE > 512 * 1024) |
1835 | batch = (512 * 1024) / PAGE_SIZE; | |
e7c8d5c9 CL |
1836 | batch /= 4; /* We effectively *= 4 below */ |
1837 | if (batch < 1) | |
1838 | batch = 1; | |
1839 | ||
1840 | /* | |
0ceaacc9 NP |
1841 | * Clamp the batch to a 2^n - 1 value. Having a power |
1842 | * of 2 value was found to be more likely to have | |
1843 | * suboptimal cache aliasing properties in some cases. | |
e7c8d5c9 | 1844 | * |
0ceaacc9 NP |
1845 | * For example if 2 tasks are alternately allocating |
1846 | * batches of pages, one task can end up with a lot | |
1847 | * of pages of one half of the possible page colors | |
1848 | * and the other with pages of the other colors. | |
e7c8d5c9 | 1849 | */ |
0ceaacc9 | 1850 | batch = (1 << (fls(batch + batch/2)-1)) - 1; |
ba56e91c | 1851 | |
e7c8d5c9 CL |
1852 | return batch; |
1853 | } | |
1854 | ||
2caaad41 CL |
1855 | inline void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
1856 | { | |
1857 | struct per_cpu_pages *pcp; | |
1858 | ||
1c6fe946 MD |
1859 | memset(p, 0, sizeof(*p)); |
1860 | ||
2caaad41 CL |
1861 | pcp = &p->pcp[0]; /* hot */ |
1862 | pcp->count = 0; | |
2caaad41 CL |
1863 | pcp->high = 6 * batch; |
1864 | pcp->batch = max(1UL, 1 * batch); | |
1865 | INIT_LIST_HEAD(&pcp->list); | |
1866 | ||
1867 | pcp = &p->pcp[1]; /* cold*/ | |
1868 | pcp->count = 0; | |
2caaad41 | 1869 | pcp->high = 2 * batch; |
e46a5e28 | 1870 | pcp->batch = max(1UL, batch/2); |
2caaad41 CL |
1871 | INIT_LIST_HEAD(&pcp->list); |
1872 | } | |
1873 | ||
8ad4b1fb RS |
1874 | /* |
1875 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist | |
1876 | * to the value high for the pageset p. | |
1877 | */ | |
1878 | ||
1879 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, | |
1880 | unsigned long high) | |
1881 | { | |
1882 | struct per_cpu_pages *pcp; | |
1883 | ||
1884 | pcp = &p->pcp[0]; /* hot list */ | |
1885 | pcp->high = high; | |
1886 | pcp->batch = max(1UL, high/4); | |
1887 | if ((high/4) > (PAGE_SHIFT * 8)) | |
1888 | pcp->batch = PAGE_SHIFT * 8; | |
1889 | } | |
1890 | ||
1891 | ||
e7c8d5c9 CL |
1892 | #ifdef CONFIG_NUMA |
1893 | /* | |
2caaad41 CL |
1894 | * Boot pageset table. One per cpu which is going to be used for all |
1895 | * zones and all nodes. The parameters will be set in such a way | |
1896 | * that an item put on a list will immediately be handed over to | |
1897 | * the buddy list. This is safe since pageset manipulation is done | |
1898 | * with interrupts disabled. | |
1899 | * | |
1900 | * Some NUMA counter updates may also be caught by the boot pagesets. | |
b7c84c6a CL |
1901 | * |
1902 | * The boot_pagesets must be kept even after bootup is complete for | |
1903 | * unused processors and/or zones. They do play a role for bootstrapping | |
1904 | * hotplugged processors. | |
1905 | * | |
1906 | * zoneinfo_show() and maybe other functions do | |
1907 | * not check if the processor is online before following the pageset pointer. | |
1908 | * Other parts of the kernel may not check if the zone is available. | |
2caaad41 | 1909 | */ |
88a2a4ac | 1910 | static struct per_cpu_pageset boot_pageset[NR_CPUS]; |
2caaad41 CL |
1911 | |
1912 | /* | |
1913 | * Dynamically allocate memory for the | |
e7c8d5c9 CL |
1914 | * per cpu pageset array in struct zone. |
1915 | */ | |
6292d9aa | 1916 | static int __cpuinit process_zones(int cpu) |
e7c8d5c9 CL |
1917 | { |
1918 | struct zone *zone, *dzone; | |
e7c8d5c9 CL |
1919 | |
1920 | for_each_zone(zone) { | |
e7c8d5c9 | 1921 | |
23316bc8 | 1922 | zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset), |
e7c8d5c9 | 1923 | GFP_KERNEL, cpu_to_node(cpu)); |
23316bc8 | 1924 | if (!zone_pcp(zone, cpu)) |
e7c8d5c9 | 1925 | goto bad; |
e7c8d5c9 | 1926 | |
23316bc8 | 1927 | setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone)); |
8ad4b1fb RS |
1928 | |
1929 | if (percpu_pagelist_fraction) | |
1930 | setup_pagelist_highmark(zone_pcp(zone, cpu), | |
1931 | (zone->present_pages / percpu_pagelist_fraction)); | |
e7c8d5c9 CL |
1932 | } |
1933 | ||
1934 | return 0; | |
1935 | bad: | |
1936 | for_each_zone(dzone) { | |
1937 | if (dzone == zone) | |
1938 | break; | |
23316bc8 NP |
1939 | kfree(zone_pcp(dzone, cpu)); |
1940 | zone_pcp(dzone, cpu) = NULL; | |
e7c8d5c9 CL |
1941 | } |
1942 | return -ENOMEM; | |
1943 | } | |
1944 | ||
1945 | static inline void free_zone_pagesets(int cpu) | |
1946 | { | |
e7c8d5c9 CL |
1947 | struct zone *zone; |
1948 | ||
1949 | for_each_zone(zone) { | |
1950 | struct per_cpu_pageset *pset = zone_pcp(zone, cpu); | |
1951 | ||
1952 | zone_pcp(zone, cpu) = NULL; | |
1953 | kfree(pset); | |
1954 | } | |
e7c8d5c9 CL |
1955 | } |
1956 | ||
6292d9aa | 1957 | static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb, |
e7c8d5c9 CL |
1958 | unsigned long action, |
1959 | void *hcpu) | |
1960 | { | |
1961 | int cpu = (long)hcpu; | |
1962 | int ret = NOTIFY_OK; | |
1963 | ||
1964 | switch (action) { | |
1965 | case CPU_UP_PREPARE: | |
1966 | if (process_zones(cpu)) | |
1967 | ret = NOTIFY_BAD; | |
1968 | break; | |
b0d41693 | 1969 | case CPU_UP_CANCELED: |
e7c8d5c9 CL |
1970 | case CPU_DEAD: |
1971 | free_zone_pagesets(cpu); | |
1972 | break; | |
e7c8d5c9 CL |
1973 | default: |
1974 | break; | |
1975 | } | |
1976 | return ret; | |
1977 | } | |
1978 | ||
1979 | static struct notifier_block pageset_notifier = | |
1980 | { &pageset_cpuup_callback, NULL, 0 }; | |
1981 | ||
78d9955b | 1982 | void __init setup_per_cpu_pageset(void) |
e7c8d5c9 CL |
1983 | { |
1984 | int err; | |
1985 | ||
1986 | /* Initialize per_cpu_pageset for cpu 0. | |
1987 | * A cpuup callback will do this for every cpu | |
1988 | * as it comes online | |
1989 | */ | |
1990 | err = process_zones(smp_processor_id()); | |
1991 | BUG_ON(err); | |
1992 | register_cpu_notifier(&pageset_notifier); | |
1993 | } | |
1994 | ||
1995 | #endif | |
1996 | ||
c09b4240 | 1997 | static __meminit |
ed8ece2e DH |
1998 | void zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
1999 | { | |
2000 | int i; | |
2001 | struct pglist_data *pgdat = zone->zone_pgdat; | |
2002 | ||
2003 | /* | |
2004 | * The per-page waitqueue mechanism uses hashed waitqueues | |
2005 | * per zone. | |
2006 | */ | |
2007 | zone->wait_table_size = wait_table_size(zone_size_pages); | |
2008 | zone->wait_table_bits = wait_table_bits(zone->wait_table_size); | |
2009 | zone->wait_table = (wait_queue_head_t *) | |
2010 | alloc_bootmem_node(pgdat, zone->wait_table_size | |
2011 | * sizeof(wait_queue_head_t)); | |
2012 | ||
2013 | for(i = 0; i < zone->wait_table_size; ++i) | |
2014 | init_waitqueue_head(zone->wait_table + i); | |
2015 | } | |
2016 | ||
c09b4240 | 2017 | static __meminit void zone_pcp_init(struct zone *zone) |
ed8ece2e DH |
2018 | { |
2019 | int cpu; | |
2020 | unsigned long batch = zone_batchsize(zone); | |
2021 | ||
2022 | for (cpu = 0; cpu < NR_CPUS; cpu++) { | |
2023 | #ifdef CONFIG_NUMA | |
2024 | /* Early boot. Slab allocator not functional yet */ | |
23316bc8 | 2025 | zone_pcp(zone, cpu) = &boot_pageset[cpu]; |
ed8ece2e DH |
2026 | setup_pageset(&boot_pageset[cpu],0); |
2027 | #else | |
2028 | setup_pageset(zone_pcp(zone,cpu), batch); | |
2029 | #endif | |
2030 | } | |
f5335c0f AB |
2031 | if (zone->present_pages) |
2032 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%lu\n", | |
2033 | zone->name, zone->present_pages, batch); | |
ed8ece2e DH |
2034 | } |
2035 | ||
c09b4240 | 2036 | static __meminit void init_currently_empty_zone(struct zone *zone, |
ed8ece2e DH |
2037 | unsigned long zone_start_pfn, unsigned long size) |
2038 | { | |
2039 | struct pglist_data *pgdat = zone->zone_pgdat; | |
2040 | ||
2041 | zone_wait_table_init(zone, size); | |
2042 | pgdat->nr_zones = zone_idx(zone) + 1; | |
2043 | ||
ed8ece2e DH |
2044 | zone->zone_start_pfn = zone_start_pfn; |
2045 | ||
2046 | memmap_init(size, pgdat->node_id, zone_idx(zone), zone_start_pfn); | |
2047 | ||
2048 | zone_init_free_lists(pgdat, zone, zone->spanned_pages); | |
2049 | } | |
2050 | ||
1da177e4 LT |
2051 | /* |
2052 | * Set up the zone data structures: | |
2053 | * - mark all pages reserved | |
2054 | * - mark all memory queues empty | |
2055 | * - clear the memory bitmaps | |
2056 | */ | |
2057 | static void __init free_area_init_core(struct pglist_data *pgdat, | |
2058 | unsigned long *zones_size, unsigned long *zholes_size) | |
2059 | { | |
ed8ece2e DH |
2060 | unsigned long j; |
2061 | int nid = pgdat->node_id; | |
1da177e4 LT |
2062 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
2063 | ||
208d54e5 | 2064 | pgdat_resize_init(pgdat); |
1da177e4 LT |
2065 | pgdat->nr_zones = 0; |
2066 | init_waitqueue_head(&pgdat->kswapd_wait); | |
2067 | pgdat->kswapd_max_order = 0; | |
2068 | ||
2069 | for (j = 0; j < MAX_NR_ZONES; j++) { | |
2070 | struct zone *zone = pgdat->node_zones + j; | |
2071 | unsigned long size, realsize; | |
1da177e4 | 2072 | |
1da177e4 LT |
2073 | realsize = size = zones_size[j]; |
2074 | if (zholes_size) | |
2075 | realsize -= zholes_size[j]; | |
2076 | ||
a2f1b424 | 2077 | if (j < ZONE_HIGHMEM) |
1da177e4 LT |
2078 | nr_kernel_pages += realsize; |
2079 | nr_all_pages += realsize; | |
2080 | ||
2081 | zone->spanned_pages = size; | |
2082 | zone->present_pages = realsize; | |
2083 | zone->name = zone_names[j]; | |
2084 | spin_lock_init(&zone->lock); | |
2085 | spin_lock_init(&zone->lru_lock); | |
bdc8cb98 | 2086 | zone_seqlock_init(zone); |
1da177e4 LT |
2087 | zone->zone_pgdat = pgdat; |
2088 | zone->free_pages = 0; | |
2089 | ||
2090 | zone->temp_priority = zone->prev_priority = DEF_PRIORITY; | |
2091 | ||
ed8ece2e | 2092 | zone_pcp_init(zone); |
1da177e4 LT |
2093 | INIT_LIST_HEAD(&zone->active_list); |
2094 | INIT_LIST_HEAD(&zone->inactive_list); | |
2095 | zone->nr_scan_active = 0; | |
2096 | zone->nr_scan_inactive = 0; | |
2097 | zone->nr_active = 0; | |
2098 | zone->nr_inactive = 0; | |
53e9a615 | 2099 | atomic_set(&zone->reclaim_in_progress, 0); |
1da177e4 LT |
2100 | if (!size) |
2101 | continue; | |
2102 | ||
d41dee36 | 2103 | zonetable_add(zone, nid, j, zone_start_pfn, size); |
ed8ece2e | 2104 | init_currently_empty_zone(zone, zone_start_pfn, size); |
1da177e4 | 2105 | zone_start_pfn += size; |
1da177e4 LT |
2106 | } |
2107 | } | |
2108 | ||
2109 | static void __init alloc_node_mem_map(struct pglist_data *pgdat) | |
2110 | { | |
1da177e4 LT |
2111 | /* Skip empty nodes */ |
2112 | if (!pgdat->node_spanned_pages) | |
2113 | return; | |
2114 | ||
d41dee36 | 2115 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
1da177e4 LT |
2116 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
2117 | if (!pgdat->node_mem_map) { | |
d41dee36 AW |
2118 | unsigned long size; |
2119 | struct page *map; | |
2120 | ||
1da177e4 | 2121 | size = (pgdat->node_spanned_pages + 1) * sizeof(struct page); |
6f167ec7 DH |
2122 | map = alloc_remap(pgdat->node_id, size); |
2123 | if (!map) | |
2124 | map = alloc_bootmem_node(pgdat, size); | |
2125 | pgdat->node_mem_map = map; | |
1da177e4 | 2126 | } |
d41dee36 | 2127 | #ifdef CONFIG_FLATMEM |
1da177e4 LT |
2128 | /* |
2129 | * With no DISCONTIG, the global mem_map is just set as node 0's | |
2130 | */ | |
2131 | if (pgdat == NODE_DATA(0)) | |
2132 | mem_map = NODE_DATA(0)->node_mem_map; | |
2133 | #endif | |
d41dee36 | 2134 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
1da177e4 LT |
2135 | } |
2136 | ||
2137 | void __init free_area_init_node(int nid, struct pglist_data *pgdat, | |
2138 | unsigned long *zones_size, unsigned long node_start_pfn, | |
2139 | unsigned long *zholes_size) | |
2140 | { | |
2141 | pgdat->node_id = nid; | |
2142 | pgdat->node_start_pfn = node_start_pfn; | |
2143 | calculate_zone_totalpages(pgdat, zones_size, zholes_size); | |
2144 | ||
2145 | alloc_node_mem_map(pgdat); | |
2146 | ||
2147 | free_area_init_core(pgdat, zones_size, zholes_size); | |
2148 | } | |
2149 | ||
93b7504e | 2150 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
1da177e4 LT |
2151 | static bootmem_data_t contig_bootmem_data; |
2152 | struct pglist_data contig_page_data = { .bdata = &contig_bootmem_data }; | |
2153 | ||
2154 | EXPORT_SYMBOL(contig_page_data); | |
93b7504e | 2155 | #endif |
1da177e4 LT |
2156 | |
2157 | void __init free_area_init(unsigned long *zones_size) | |
2158 | { | |
93b7504e | 2159 | free_area_init_node(0, NODE_DATA(0), zones_size, |
1da177e4 LT |
2160 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
2161 | } | |
1da177e4 LT |
2162 | |
2163 | #ifdef CONFIG_PROC_FS | |
2164 | ||
2165 | #include <linux/seq_file.h> | |
2166 | ||
2167 | static void *frag_start(struct seq_file *m, loff_t *pos) | |
2168 | { | |
2169 | pg_data_t *pgdat; | |
2170 | loff_t node = *pos; | |
ae0f15fb KH |
2171 | for (pgdat = first_online_pgdat(); |
2172 | pgdat && node; | |
2173 | pgdat = next_online_pgdat(pgdat)) | |
1da177e4 LT |
2174 | --node; |
2175 | ||
2176 | return pgdat; | |
2177 | } | |
2178 | ||
2179 | static void *frag_next(struct seq_file *m, void *arg, loff_t *pos) | |
2180 | { | |
2181 | pg_data_t *pgdat = (pg_data_t *)arg; | |
2182 | ||
2183 | (*pos)++; | |
ae0f15fb | 2184 | return next_online_pgdat(pgdat); |
1da177e4 LT |
2185 | } |
2186 | ||
2187 | static void frag_stop(struct seq_file *m, void *arg) | |
2188 | { | |
2189 | } | |
2190 | ||
2191 | /* | |
2192 | * This walks the free areas for each zone. | |
2193 | */ | |
2194 | static int frag_show(struct seq_file *m, void *arg) | |
2195 | { | |
2196 | pg_data_t *pgdat = (pg_data_t *)arg; | |
2197 | struct zone *zone; | |
2198 | struct zone *node_zones = pgdat->node_zones; | |
2199 | unsigned long flags; | |
2200 | int order; | |
2201 | ||
2202 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; ++zone) { | |
f3fe6512 | 2203 | if (!populated_zone(zone)) |
1da177e4 LT |
2204 | continue; |
2205 | ||
2206 | spin_lock_irqsave(&zone->lock, flags); | |
2207 | seq_printf(m, "Node %d, zone %8s ", pgdat->node_id, zone->name); | |
2208 | for (order = 0; order < MAX_ORDER; ++order) | |
2209 | seq_printf(m, "%6lu ", zone->free_area[order].nr_free); | |
2210 | spin_unlock_irqrestore(&zone->lock, flags); | |
2211 | seq_putc(m, '\n'); | |
2212 | } | |
2213 | return 0; | |
2214 | } | |
2215 | ||
2216 | struct seq_operations fragmentation_op = { | |
2217 | .start = frag_start, | |
2218 | .next = frag_next, | |
2219 | .stop = frag_stop, | |
2220 | .show = frag_show, | |
2221 | }; | |
2222 | ||
295ab934 ND |
2223 | /* |
2224 | * Output information about zones in @pgdat. | |
2225 | */ | |
2226 | static int zoneinfo_show(struct seq_file *m, void *arg) | |
2227 | { | |
2228 | pg_data_t *pgdat = arg; | |
2229 | struct zone *zone; | |
2230 | struct zone *node_zones = pgdat->node_zones; | |
2231 | unsigned long flags; | |
2232 | ||
2233 | for (zone = node_zones; zone - node_zones < MAX_NR_ZONES; zone++) { | |
2234 | int i; | |
2235 | ||
f3fe6512 | 2236 | if (!populated_zone(zone)) |
295ab934 ND |
2237 | continue; |
2238 | ||
2239 | spin_lock_irqsave(&zone->lock, flags); | |
2240 | seq_printf(m, "Node %d, zone %8s", pgdat->node_id, zone->name); | |
2241 | seq_printf(m, | |
2242 | "\n pages free %lu" | |
2243 | "\n min %lu" | |
2244 | "\n low %lu" | |
2245 | "\n high %lu" | |
2246 | "\n active %lu" | |
2247 | "\n inactive %lu" | |
2248 | "\n scanned %lu (a: %lu i: %lu)" | |
2249 | "\n spanned %lu" | |
2250 | "\n present %lu", | |
2251 | zone->free_pages, | |
2252 | zone->pages_min, | |
2253 | zone->pages_low, | |
2254 | zone->pages_high, | |
2255 | zone->nr_active, | |
2256 | zone->nr_inactive, | |
2257 | zone->pages_scanned, | |
2258 | zone->nr_scan_active, zone->nr_scan_inactive, | |
2259 | zone->spanned_pages, | |
2260 | zone->present_pages); | |
2261 | seq_printf(m, | |
2262 | "\n protection: (%lu", | |
2263 | zone->lowmem_reserve[0]); | |
2264 | for (i = 1; i < ARRAY_SIZE(zone->lowmem_reserve); i++) | |
2265 | seq_printf(m, ", %lu", zone->lowmem_reserve[i]); | |
2266 | seq_printf(m, | |
2267 | ")" | |
2268 | "\n pagesets"); | |
23316bc8 | 2269 | for_each_online_cpu(i) { |
295ab934 ND |
2270 | struct per_cpu_pageset *pageset; |
2271 | int j; | |
2272 | ||
e7c8d5c9 | 2273 | pageset = zone_pcp(zone, i); |
295ab934 ND |
2274 | for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) { |
2275 | if (pageset->pcp[j].count) | |
2276 | break; | |
2277 | } | |
2278 | if (j == ARRAY_SIZE(pageset->pcp)) | |
2279 | continue; | |
2280 | for (j = 0; j < ARRAY_SIZE(pageset->pcp); j++) { | |
2281 | seq_printf(m, | |
2282 | "\n cpu: %i pcp: %i" | |
2283 | "\n count: %i" | |
295ab934 ND |
2284 | "\n high: %i" |
2285 | "\n batch: %i", | |
2286 | i, j, | |
2287 | pageset->pcp[j].count, | |
295ab934 ND |
2288 | pageset->pcp[j].high, |
2289 | pageset->pcp[j].batch); | |
2290 | } | |
2291 | #ifdef CONFIG_NUMA | |
2292 | seq_printf(m, | |
2293 | "\n numa_hit: %lu" | |
2294 | "\n numa_miss: %lu" | |
2295 | "\n numa_foreign: %lu" | |
2296 | "\n interleave_hit: %lu" | |
2297 | "\n local_node: %lu" | |
2298 | "\n other_node: %lu", | |
2299 | pageset->numa_hit, | |
2300 | pageset->numa_miss, | |
2301 | pageset->numa_foreign, | |
2302 | pageset->interleave_hit, | |
2303 | pageset->local_node, | |
2304 | pageset->other_node); | |
2305 | #endif | |
2306 | } | |
2307 | seq_printf(m, | |
2308 | "\n all_unreclaimable: %u" | |
2309 | "\n prev_priority: %i" | |
2310 | "\n temp_priority: %i" | |
2311 | "\n start_pfn: %lu", | |
2312 | zone->all_unreclaimable, | |
2313 | zone->prev_priority, | |
2314 | zone->temp_priority, | |
2315 | zone->zone_start_pfn); | |
2316 | spin_unlock_irqrestore(&zone->lock, flags); | |
2317 | seq_putc(m, '\n'); | |
2318 | } | |
2319 | return 0; | |
2320 | } | |
2321 | ||
2322 | struct seq_operations zoneinfo_op = { | |
2323 | .start = frag_start, /* iterate over all zones. The same as in | |
2324 | * fragmentation. */ | |
2325 | .next = frag_next, | |
2326 | .stop = frag_stop, | |
2327 | .show = zoneinfo_show, | |
2328 | }; | |
2329 | ||
1da177e4 LT |
2330 | static char *vmstat_text[] = { |
2331 | "nr_dirty", | |
2332 | "nr_writeback", | |
2333 | "nr_unstable", | |
2334 | "nr_page_table_pages", | |
2335 | "nr_mapped", | |
2336 | "nr_slab", | |
2337 | ||
2338 | "pgpgin", | |
2339 | "pgpgout", | |
2340 | "pswpin", | |
2341 | "pswpout", | |
1da177e4 | 2342 | |
9328b8fa | 2343 | "pgalloc_high", |
1da177e4 | 2344 | "pgalloc_normal", |
9328b8fa | 2345 | "pgalloc_dma32", |
1da177e4 | 2346 | "pgalloc_dma", |
9328b8fa | 2347 | |
1da177e4 LT |
2348 | "pgfree", |
2349 | "pgactivate", | |
2350 | "pgdeactivate", | |
2351 | ||
2352 | "pgfault", | |
2353 | "pgmajfault", | |
9328b8fa | 2354 | |
1da177e4 LT |
2355 | "pgrefill_high", |
2356 | "pgrefill_normal", | |
9328b8fa | 2357 | "pgrefill_dma32", |
1da177e4 LT |
2358 | "pgrefill_dma", |
2359 | ||
2360 | "pgsteal_high", | |
2361 | "pgsteal_normal", | |
9328b8fa | 2362 | "pgsteal_dma32", |
1da177e4 | 2363 | "pgsteal_dma", |
9328b8fa | 2364 | |
1da177e4 LT |
2365 | "pgscan_kswapd_high", |
2366 | "pgscan_kswapd_normal", | |
9328b8fa | 2367 | "pgscan_kswapd_dma32", |
1da177e4 | 2368 | "pgscan_kswapd_dma", |
9328b8fa | 2369 | |
1da177e4 LT |
2370 | "pgscan_direct_high", |
2371 | "pgscan_direct_normal", | |
9328b8fa | 2372 | "pgscan_direct_dma32", |
1da177e4 | 2373 | "pgscan_direct_dma", |
1da177e4 | 2374 | |
9328b8fa | 2375 | "pginodesteal", |
1da177e4 LT |
2376 | "slabs_scanned", |
2377 | "kswapd_steal", | |
2378 | "kswapd_inodesteal", | |
2379 | "pageoutrun", | |
2380 | "allocstall", | |
2381 | ||
2382 | "pgrotated", | |
edfbe2b0 | 2383 | "nr_bounce", |
1da177e4 LT |
2384 | }; |
2385 | ||
2386 | static void *vmstat_start(struct seq_file *m, loff_t *pos) | |
2387 | { | |
2388 | struct page_state *ps; | |
2389 | ||
2390 | if (*pos >= ARRAY_SIZE(vmstat_text)) | |
2391 | return NULL; | |
2392 | ||
2393 | ps = kmalloc(sizeof(*ps), GFP_KERNEL); | |
2394 | m->private = ps; | |
2395 | if (!ps) | |
2396 | return ERR_PTR(-ENOMEM); | |
2397 | get_full_page_state(ps); | |
2398 | ps->pgpgin /= 2; /* sectors -> kbytes */ | |
2399 | ps->pgpgout /= 2; | |
2400 | return (unsigned long *)ps + *pos; | |
2401 | } | |
2402 | ||
2403 | static void *vmstat_next(struct seq_file *m, void *arg, loff_t *pos) | |
2404 | { | |
2405 | (*pos)++; | |
2406 | if (*pos >= ARRAY_SIZE(vmstat_text)) | |
2407 | return NULL; | |
2408 | return (unsigned long *)m->private + *pos; | |
2409 | } | |
2410 | ||
2411 | static int vmstat_show(struct seq_file *m, void *arg) | |
2412 | { | |
2413 | unsigned long *l = arg; | |
2414 | unsigned long off = l - (unsigned long *)m->private; | |
2415 | ||
2416 | seq_printf(m, "%s %lu\n", vmstat_text[off], *l); | |
2417 | return 0; | |
2418 | } | |
2419 | ||
2420 | static void vmstat_stop(struct seq_file *m, void *arg) | |
2421 | { | |
2422 | kfree(m->private); | |
2423 | m->private = NULL; | |
2424 | } | |
2425 | ||
2426 | struct seq_operations vmstat_op = { | |
2427 | .start = vmstat_start, | |
2428 | .next = vmstat_next, | |
2429 | .stop = vmstat_stop, | |
2430 | .show = vmstat_show, | |
2431 | }; | |
2432 | ||
2433 | #endif /* CONFIG_PROC_FS */ | |
2434 | ||
2435 | #ifdef CONFIG_HOTPLUG_CPU | |
2436 | static int page_alloc_cpu_notify(struct notifier_block *self, | |
2437 | unsigned long action, void *hcpu) | |
2438 | { | |
2439 | int cpu = (unsigned long)hcpu; | |
2440 | long *count; | |
2441 | unsigned long *src, *dest; | |
2442 | ||
2443 | if (action == CPU_DEAD) { | |
2444 | int i; | |
2445 | ||
2446 | /* Drain local pagecache count. */ | |
2447 | count = &per_cpu(nr_pagecache_local, cpu); | |
2448 | atomic_add(*count, &nr_pagecache); | |
2449 | *count = 0; | |
2450 | local_irq_disable(); | |
2451 | __drain_pages(cpu); | |
2452 | ||
2453 | /* Add dead cpu's page_states to our own. */ | |
2454 | dest = (unsigned long *)&__get_cpu_var(page_states); | |
2455 | src = (unsigned long *)&per_cpu(page_states, cpu); | |
2456 | ||
2457 | for (i = 0; i < sizeof(struct page_state)/sizeof(unsigned long); | |
2458 | i++) { | |
2459 | dest[i] += src[i]; | |
2460 | src[i] = 0; | |
2461 | } | |
2462 | ||
2463 | local_irq_enable(); | |
2464 | } | |
2465 | return NOTIFY_OK; | |
2466 | } | |
2467 | #endif /* CONFIG_HOTPLUG_CPU */ | |
2468 | ||
2469 | void __init page_alloc_init(void) | |
2470 | { | |
2471 | hotcpu_notifier(page_alloc_cpu_notify, 0); | |
2472 | } | |
2473 | ||
2474 | /* | |
2475 | * setup_per_zone_lowmem_reserve - called whenever | |
2476 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone | |
2477 | * has a correct pages reserved value, so an adequate number of | |
2478 | * pages are left in the zone after a successful __alloc_pages(). | |
2479 | */ | |
2480 | static void setup_per_zone_lowmem_reserve(void) | |
2481 | { | |
2482 | struct pglist_data *pgdat; | |
2483 | int j, idx; | |
2484 | ||
ec936fc5 | 2485 | for_each_online_pgdat(pgdat) { |
1da177e4 LT |
2486 | for (j = 0; j < MAX_NR_ZONES; j++) { |
2487 | struct zone *zone = pgdat->node_zones + j; | |
2488 | unsigned long present_pages = zone->present_pages; | |
2489 | ||
2490 | zone->lowmem_reserve[j] = 0; | |
2491 | ||
2492 | for (idx = j-1; idx >= 0; idx--) { | |
2493 | struct zone *lower_zone; | |
2494 | ||
2495 | if (sysctl_lowmem_reserve_ratio[idx] < 1) | |
2496 | sysctl_lowmem_reserve_ratio[idx] = 1; | |
2497 | ||
2498 | lower_zone = pgdat->node_zones + idx; | |
2499 | lower_zone->lowmem_reserve[j] = present_pages / | |
2500 | sysctl_lowmem_reserve_ratio[idx]; | |
2501 | present_pages += lower_zone->present_pages; | |
2502 | } | |
2503 | } | |
2504 | } | |
2505 | } | |
2506 | ||
2507 | /* | |
2508 | * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures | |
2509 | * that the pages_{min,low,high} values for each zone are set correctly | |
2510 | * with respect to min_free_kbytes. | |
2511 | */ | |
3947be19 | 2512 | void setup_per_zone_pages_min(void) |
1da177e4 LT |
2513 | { |
2514 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); | |
2515 | unsigned long lowmem_pages = 0; | |
2516 | struct zone *zone; | |
2517 | unsigned long flags; | |
2518 | ||
2519 | /* Calculate total number of !ZONE_HIGHMEM pages */ | |
2520 | for_each_zone(zone) { | |
2521 | if (!is_highmem(zone)) | |
2522 | lowmem_pages += zone->present_pages; | |
2523 | } | |
2524 | ||
2525 | for_each_zone(zone) { | |
669ed175 | 2526 | unsigned long tmp; |
1da177e4 | 2527 | spin_lock_irqsave(&zone->lru_lock, flags); |
669ed175 | 2528 | tmp = (pages_min * zone->present_pages) / lowmem_pages; |
1da177e4 LT |
2529 | if (is_highmem(zone)) { |
2530 | /* | |
669ed175 NP |
2531 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
2532 | * need highmem pages, so cap pages_min to a small | |
2533 | * value here. | |
2534 | * | |
2535 | * The (pages_high-pages_low) and (pages_low-pages_min) | |
2536 | * deltas controls asynch page reclaim, and so should | |
2537 | * not be capped for highmem. | |
1da177e4 LT |
2538 | */ |
2539 | int min_pages; | |
2540 | ||
2541 | min_pages = zone->present_pages / 1024; | |
2542 | if (min_pages < SWAP_CLUSTER_MAX) | |
2543 | min_pages = SWAP_CLUSTER_MAX; | |
2544 | if (min_pages > 128) | |
2545 | min_pages = 128; | |
2546 | zone->pages_min = min_pages; | |
2547 | } else { | |
669ed175 NP |
2548 | /* |
2549 | * If it's a lowmem zone, reserve a number of pages | |
1da177e4 LT |
2550 | * proportionate to the zone's size. |
2551 | */ | |
669ed175 | 2552 | zone->pages_min = tmp; |
1da177e4 LT |
2553 | } |
2554 | ||
669ed175 NP |
2555 | zone->pages_low = zone->pages_min + tmp / 4; |
2556 | zone->pages_high = zone->pages_min + tmp / 2; | |
1da177e4 LT |
2557 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
2558 | } | |
2559 | } | |
2560 | ||
2561 | /* | |
2562 | * Initialise min_free_kbytes. | |
2563 | * | |
2564 | * For small machines we want it small (128k min). For large machines | |
2565 | * we want it large (64MB max). But it is not linear, because network | |
2566 | * bandwidth does not increase linearly with machine size. We use | |
2567 | * | |
2568 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: | |
2569 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) | |
2570 | * | |
2571 | * which yields | |
2572 | * | |
2573 | * 16MB: 512k | |
2574 | * 32MB: 724k | |
2575 | * 64MB: 1024k | |
2576 | * 128MB: 1448k | |
2577 | * 256MB: 2048k | |
2578 | * 512MB: 2896k | |
2579 | * 1024MB: 4096k | |
2580 | * 2048MB: 5792k | |
2581 | * 4096MB: 8192k | |
2582 | * 8192MB: 11584k | |
2583 | * 16384MB: 16384k | |
2584 | */ | |
2585 | static int __init init_per_zone_pages_min(void) | |
2586 | { | |
2587 | unsigned long lowmem_kbytes; | |
2588 | ||
2589 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); | |
2590 | ||
2591 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); | |
2592 | if (min_free_kbytes < 128) | |
2593 | min_free_kbytes = 128; | |
2594 | if (min_free_kbytes > 65536) | |
2595 | min_free_kbytes = 65536; | |
2596 | setup_per_zone_pages_min(); | |
2597 | setup_per_zone_lowmem_reserve(); | |
2598 | return 0; | |
2599 | } | |
2600 | module_init(init_per_zone_pages_min) | |
2601 | ||
2602 | /* | |
2603 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so | |
2604 | * that we can call two helper functions whenever min_free_kbytes | |
2605 | * changes. | |
2606 | */ | |
2607 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, | |
2608 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2609 | { | |
2610 | proc_dointvec(table, write, file, buffer, length, ppos); | |
2611 | setup_per_zone_pages_min(); | |
2612 | return 0; | |
2613 | } | |
2614 | ||
2615 | /* | |
2616 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around | |
2617 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() | |
2618 | * whenever sysctl_lowmem_reserve_ratio changes. | |
2619 | * | |
2620 | * The reserve ratio obviously has absolutely no relation with the | |
2621 | * pages_min watermarks. The lowmem reserve ratio can only make sense | |
2622 | * if in function of the boot time zone sizes. | |
2623 | */ | |
2624 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, | |
2625 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2626 | { | |
2627 | proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2628 | setup_per_zone_lowmem_reserve(); | |
2629 | return 0; | |
2630 | } | |
2631 | ||
8ad4b1fb RS |
2632 | /* |
2633 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each | |
2634 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist | |
2635 | * can have before it gets flushed back to buddy allocator. | |
2636 | */ | |
2637 | ||
2638 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, | |
2639 | struct file *file, void __user *buffer, size_t *length, loff_t *ppos) | |
2640 | { | |
2641 | struct zone *zone; | |
2642 | unsigned int cpu; | |
2643 | int ret; | |
2644 | ||
2645 | ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos); | |
2646 | if (!write || (ret == -EINVAL)) | |
2647 | return ret; | |
2648 | for_each_zone(zone) { | |
2649 | for_each_online_cpu(cpu) { | |
2650 | unsigned long high; | |
2651 | high = zone->present_pages / percpu_pagelist_fraction; | |
2652 | setup_pagelist_highmark(zone_pcp(zone, cpu), high); | |
2653 | } | |
2654 | } | |
2655 | return 0; | |
2656 | } | |
2657 | ||
1da177e4 LT |
2658 | __initdata int hashdist = HASHDIST_DEFAULT; |
2659 | ||
2660 | #ifdef CONFIG_NUMA | |
2661 | static int __init set_hashdist(char *str) | |
2662 | { | |
2663 | if (!str) | |
2664 | return 0; | |
2665 | hashdist = simple_strtoul(str, &str, 0); | |
2666 | return 1; | |
2667 | } | |
2668 | __setup("hashdist=", set_hashdist); | |
2669 | #endif | |
2670 | ||
2671 | /* | |
2672 | * allocate a large system hash table from bootmem | |
2673 | * - it is assumed that the hash table must contain an exact power-of-2 | |
2674 | * quantity of entries | |
2675 | * - limit is the number of hash buckets, not the total allocation size | |
2676 | */ | |
2677 | void *__init alloc_large_system_hash(const char *tablename, | |
2678 | unsigned long bucketsize, | |
2679 | unsigned long numentries, | |
2680 | int scale, | |
2681 | int flags, | |
2682 | unsigned int *_hash_shift, | |
2683 | unsigned int *_hash_mask, | |
2684 | unsigned long limit) | |
2685 | { | |
2686 | unsigned long long max = limit; | |
2687 | unsigned long log2qty, size; | |
2688 | void *table = NULL; | |
2689 | ||
2690 | /* allow the kernel cmdline to have a say */ | |
2691 | if (!numentries) { | |
2692 | /* round applicable memory size up to nearest megabyte */ | |
2693 | numentries = (flags & HASH_HIGHMEM) ? nr_all_pages : nr_kernel_pages; | |
2694 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; | |
2695 | numentries >>= 20 - PAGE_SHIFT; | |
2696 | numentries <<= 20 - PAGE_SHIFT; | |
2697 | ||
2698 | /* limit to 1 bucket per 2^scale bytes of low memory */ | |
2699 | if (scale > PAGE_SHIFT) | |
2700 | numentries >>= (scale - PAGE_SHIFT); | |
2701 | else | |
2702 | numentries <<= (PAGE_SHIFT - scale); | |
2703 | } | |
6e692ed3 | 2704 | numentries = roundup_pow_of_two(numentries); |
1da177e4 LT |
2705 | |
2706 | /* limit allocation size to 1/16 total memory by default */ | |
2707 | if (max == 0) { | |
2708 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; | |
2709 | do_div(max, bucketsize); | |
2710 | } | |
2711 | ||
2712 | if (numentries > max) | |
2713 | numentries = max; | |
2714 | ||
2715 | log2qty = long_log2(numentries); | |
2716 | ||
2717 | do { | |
2718 | size = bucketsize << log2qty; | |
2719 | if (flags & HASH_EARLY) | |
2720 | table = alloc_bootmem(size); | |
2721 | else if (hashdist) | |
2722 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); | |
2723 | else { | |
2724 | unsigned long order; | |
2725 | for (order = 0; ((1UL << order) << PAGE_SHIFT) < size; order++) | |
2726 | ; | |
2727 | table = (void*) __get_free_pages(GFP_ATOMIC, order); | |
2728 | } | |
2729 | } while (!table && size > PAGE_SIZE && --log2qty); | |
2730 | ||
2731 | if (!table) | |
2732 | panic("Failed to allocate %s hash table\n", tablename); | |
2733 | ||
2734 | printk("%s hash table entries: %d (order: %d, %lu bytes)\n", | |
2735 | tablename, | |
2736 | (1U << log2qty), | |
2737 | long_log2(size) - PAGE_SHIFT, | |
2738 | size); | |
2739 | ||
2740 | if (_hash_shift) | |
2741 | *_hash_shift = log2qty; | |
2742 | if (_hash_mask) | |
2743 | *_hash_mask = (1 << log2qty) - 1; | |
2744 | ||
2745 | return table; | |
2746 | } | |
a117e66e KH |
2747 | |
2748 | #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE | |
2749 | /* | |
2750 | * pfn <-> page translation. out-of-line version. | |
2751 | * (see asm-generic/memory_model.h) | |
2752 | */ | |
2753 | #if defined(CONFIG_FLATMEM) | |
2754 | struct page *pfn_to_page(unsigned long pfn) | |
2755 | { | |
2756 | return mem_map + (pfn - ARCH_PFN_OFFSET); | |
2757 | } | |
2758 | unsigned long page_to_pfn(struct page *page) | |
2759 | { | |
2760 | return (page - mem_map) + ARCH_PFN_OFFSET; | |
2761 | } | |
2762 | #elif defined(CONFIG_DISCONTIGMEM) | |
2763 | struct page *pfn_to_page(unsigned long pfn) | |
2764 | { | |
2765 | int nid = arch_pfn_to_nid(pfn); | |
2766 | return NODE_DATA(nid)->node_mem_map + arch_local_page_offset(pfn,nid); | |
2767 | } | |
2768 | unsigned long page_to_pfn(struct page *page) | |
2769 | { | |
a0140c1d KH |
2770 | struct pglist_data *pgdat = NODE_DATA(page_to_nid(page)); |
2771 | return (page - pgdat->node_mem_map) + pgdat->node_start_pfn; | |
a117e66e KH |
2772 | } |
2773 | #elif defined(CONFIG_SPARSEMEM) | |
2774 | struct page *pfn_to_page(unsigned long pfn) | |
2775 | { | |
2776 | return __section_mem_map_addr(__pfn_to_section(pfn)) + pfn; | |
2777 | } | |
2778 | ||
2779 | unsigned long page_to_pfn(struct page *page) | |
2780 | { | |
2781 | long section_id = page_to_section(page); | |
2782 | return page - __section_mem_map_addr(__nr_to_section(section_id)); | |
2783 | } | |
2784 | #endif /* CONFIG_FLATMEM/DISCONTIGMME/SPARSEMEM */ | |
2785 | EXPORT_SYMBOL(pfn_to_page); | |
2786 | EXPORT_SYMBOL(page_to_pfn); | |
2787 | #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */ |